WO2024067568A1 - Recommendation-information generation method and apparatus, and device, medium and program product - Google Patents

Recommendation-information generation method and apparatus, and device, medium and program product Download PDF

Info

Publication number
WO2024067568A1
WO2024067568A1 PCT/CN2023/121495 CN2023121495W WO2024067568A1 WO 2024067568 A1 WO2024067568 A1 WO 2024067568A1 CN 2023121495 W CN2023121495 W CN 2023121495W WO 2024067568 A1 WO2024067568 A1 WO 2024067568A1
Authority
WO
WIPO (PCT)
Prior art keywords
vector
image
item
sparse feature
historical
Prior art date
Application number
PCT/CN2023/121495
Other languages
French (fr)
Chinese (zh)
Inventor
刘银星
阮涛
张政
吕晶晶
Original Assignee
北京沃东天骏信息技术有限公司
北京京东世纪贸易有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京沃东天骏信息技术有限公司, 北京京东世纪贸易有限公司 filed Critical 北京沃东天骏信息技术有限公司
Publication of WO2024067568A1 publication Critical patent/WO2024067568A1/en

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/53—Querying
    • G06F16/535—Filtering based on additional data, e.g. user or group profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00—Image coding

Definitions

  • Embodiments of the present disclosure relate to the field of computer technology, and in particular to a method, apparatus, device, medium, and program product for generating recommendation information.
  • the usual method is: first, input the multiple materials into a pre-trained user material preference model to generate a score set corresponding to the multiple materials. Then, use the score set to filter out at least one material that the target user may like from the multiple materials. Finally, recommend the at least one material to the target user.
  • Some embodiments of the present disclosure propose a recommendation information generation method, an apparatus, an electronic device, a computer-readable medium, and a computer program product to solve the technical problems mentioned in the above background technology section.
  • some embodiments of the present disclosure provide a method for generating recommendation information, comprising: obtaining a target user's historically browsed creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item, wherein the item sparse feature information set includes a creative image feature information set for the target recommended item; performing graph coding processing on the main image to obtain a main image coding vector, and performing graph coding processing on each historically browsed creative image in the historically browsed creative image sequence to generate a historical image coding vector to obtain a historical image coding vector sequence; performing graph coding processing on each user sparse feature information set
  • the feature information is encoded to generate a user sparse feature vector to obtain a user sparse feature vector set, and the sparse feature information of each item in the above-mentioned item sparse feature information set is encoded to generate an item sparse feature vector to obtain an item sparse feature vector set;
  • some embodiments of the present disclosure provide an electronic device comprising: at least one processor; and a storage device on which at least one program is stored, wherein when the at least one program is executed by the at least one processor, the at least one processor implements the method described in any implementation manner in the first aspect.
  • some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation manner in the first aspect is implemented.
  • some embodiments of the present disclosure provide a computer program product, including a computer program, which implements the method described in any implementation manner in the above-mentioned first aspect when executed by a processor.
  • FIGS. 1-2 are schematic diagrams of an application scenario of a method for generating recommendation information according to some embodiments of the present disclosure
  • FIG3 is a flow chart of some embodiments of a method for generating recommendation information according to the present disclosure
  • FIG4 is a flow chart of other embodiments of the method for generating recommendation information according to the present disclosure.
  • FIG5 is a schematic diagram of the structure of some embodiments of a device for generating recommendation information according to the present disclosure
  • FIG. 6 is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure.
  • relevant organizations or individuals shall fulfill obligations including conducting material security impact assessments, fulfilling the obligation to inform the material creators, and obtaining the authorization and consent of the material creators in advance.
  • Related recommendation information generation methods for example, first, input multiple materials into a pre-trained user material preference model to generate a score set corresponding to the multiple materials. Then, use the score set to filter out at least one material that the target user may like from the above multiple materials. Finally, recommending at least one material to the target user often has the following technical problems: only using the creative image sequence of the target user's historical browsing to train the user material preference model, so that the user material preference model can learn limited feature information, so that the user material preference model is not accurate enough. This leads to poor recommendation results.
  • some embodiments of the present disclosure propose a method and device for generating recommendation information, which generates accurate recommendation information and achieves better recommendation effect.
  • FIGS. 1-2 are schematic diagrams of an application scenario of a method for generating recommendation information according to some embodiments of the present disclosure.
  • the electronic device 101 can obtain the historical browsing creative image sequence 103 of the target user 102, the user sparse feature information set 104 of the target user 102, the main image 106 of the target recommended item 105, and the item sparse feature information set 107 of the target recommended item 105.
  • the item sparse feature information set 107 includes a creative image feature information set for the target recommended item 105.
  • the target user 102 can be "Li**”.
  • the historical browsing creative image sequence 103 can include: historical browsing creative images 1031 and historical browsing creative images 1032.
  • the user sparse feature information set 104 can be: ⁇ "gender: 1", “age: 18", “height: 184" ⁇ .
  • the target recommended item 105 can be "apple”.
  • the item sparse feature information set 107 can be: ⁇ "click rate: 0.4", “price: 5", "origin: 02" ⁇ .
  • the gender "1" can represent male.
  • the origin "02” can represent Shanghai.
  • the electronic device 101 may perform image coding processing on the main image 106 to obtain a main image coding vector 110, and perform image coding processing on each historical browsing creative image in the historical browsing creative image sequence 103 to generate a historical image coding vector to obtain a historical image coding vector sequence 108.
  • the historical image coding vector sequence 108 may include: a historical image coding vector 1081 corresponding to the historical browsing creative image 1031 and a historical image coding vector 1082 corresponding to the historical browsing creative image 1032.
  • the electronic device 101 may perform information coding on each user sparse feature information in the user sparse feature information set 104 to generate a user sparse feature vector to obtain a user sparse feature vector set 109, and perform information coding on each item sparse feature information in the item sparse feature information set 107 to generate an item sparse feature vector to obtain an item sparse feature vector set 111.
  • the user sparse feature vector set 109 may include a user sparse feature vector 1091 corresponding to "gender: 1", a user sparse feature vector 1092 corresponding to "age: 18", and a user sparse feature vector 1093 corresponding to "height: 184".
  • the item sparse feature vector set 111 may include an item sparse feature vector 1111 corresponding to "click rate: 0.4", and an item sparse feature vector 1112 corresponding to "price: 5".
  • the corresponding item sparse feature vector 1112 and the item sparse feature vector 1113 corresponding to "Origin: 02".
  • the electronic device 101 can adjust the visual preference of each historical image coding vector in the above historical image coding vector sequence 108 to obtain an adjusted historical image coding vector sequence 112.
  • the adjusted historical image coding vector sequence 112 includes: an adjusted historical image coding vector 1121 corresponding to the historical image coding vector 1081 and an adjusted historical image coding vector 1122 corresponding to the historical image coding vector 1082.
  • the electronic device 101 can generate a recommended image set 114 corresponding to the above target recommended item 105 to be pushed to the above target user 102 using the multi-head attention mechanism model 113 based on the above adjusted historical image coding vector sequence 112, the above main image coding vector 110, the above user sparse feature vector set 109 and the above item sparse feature vector set 111.
  • the above recommended image set 114 is an image subset of the creative image set corresponding to the above creative image feature information set.
  • the recommended image set 114 includes: a recommended image 1141 and a recommended image 1142 .
  • the electronic device 101 can be hardware or software.
  • the electronic device can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or it can be implemented as a single server or a single terminal device.
  • the electronic device When the electronic device is embodied as software, it can be installed in the hardware devices listed above. It can be implemented as multiple software or software modules for providing distributed services, for example, or it can be implemented as a single software or software module. No specific limitation is made here.
  • FIG. 1-FIG 2 is only for illustration purposes, and any number of electronic devices may be provided according to implementation requirements.
  • the method for generating recommendation information includes the following steps:
  • Step 301 obtaining a target user's historical browsing creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item.
  • the execution subject of the above-mentioned recommendation information generation method can obtain the target user's historical browsing creative image sequence and the user sparse features of the above-mentioned target user through a wired connection or a wireless connection.
  • the above item sparse feature information set includes a creative image feature information set for the above target recommended item.
  • the creative image feature information can be the ID identification information of the creative image of the target recommended item. There is a one-to-one correspondence between the ID identification information and the creative image.
  • the target user can be the user of the material to be recommended.
  • the material can be in the form of a picture or a video.
  • the key frames extracted from the video can be used as content recommended to the user.
  • the material can be the main picture of the item or the creative image of the item.
  • the creative image can be a pre-designed image based on the characteristics of the item.
  • the historical browsing creative image sequence can be the creative image sequence browsed by the target user on each application (Application, App) in the historical time period. For example, the current time is April 2022.
  • the historical browsing creative image sequence can be the creative image sequence browsed by users from January 2021 to April 2021.
  • the main image of the target recommended item can be the main image of the target recommended item.
  • the user sparse feature information of the target user can be the numerical information of the sparse features of the target user.
  • the sparse features of the target user may be the ID features of the target user.
  • the above-mentioned ID features may be the unique ID identifier of the user features, or the identifier of the user identity.
  • the sparse features of the target user may be one of the following: the gender ID identifier of the target user, the age ID identifier of the target user, the ID identifier of the target user, and the height ID identifier of the target user.
  • the gender ID identifier is "1", which indicates that the target user is male.
  • the gender ID identifier is "0", which indicates that the target user is female.
  • the sparse feature information of the target recommended item may be the numerical information of the sparse features of the target recommended item.
  • the sparse features of the target recommended item may be the ID features of the target recommended item.
  • the ID features may be the unique ID identifier of the item features, or the identifier of the item identity.
  • the sparse features of the target recommended item may be one of the following: the click-through rate of the target recommended item, the price of the target recommended item, and the origin of the target recommended item.
  • the origin ID identifier is "02", which indicates that the origin of the target recommended item is Shanghai.
  • the information that can be recommended to the target user may include but is not limited to one of the following: a creative image of the target recommended item, and a main image of the target recommended item.
  • Step 302 performing image coding processing on the main image to obtain a main image coding vector, and performing image coding processing on each historically browsed creative image in the historically browsed creative image sequence to generate a historical image coding vector to obtain a historical image coding vector sequence.
  • the execution subject may perform image coding processing on the main image to obtain a main image coding vector, and perform image coding processing on each historically browsed creative image in the historically browsed creative image sequence to generate a historical image coding vector to obtain a historical image coding vector sequence.
  • the main image coding vector may represent image feature information of the main image.
  • the historical image coding vector may represent image feature information of the historically browsed creative images.
  • the execution subject may input the main image into a plurality of serially connected convolutional neural networks (CNNs) to obtain a first model output result. Then, the first model output result is input into a Bert coding model to obtain a main image coding vector.
  • CNNs serially connected convolutional neural networks
  • the execution subject can input each historical browsing creative image in the historical browsing creative image sequence into a plurality of serially connected convolutional neural networks to generate a second model output result and obtain a second model output result sequence. Then, each second model output result in the second model output result sequence is input into the Bert coding model to obtain a historical image coding vector sequence.
  • using the main image encoding vector to subsequently generate multiple recommended creative image scores and recommended main image scores has a smaller amount of calculation than using the main image to subsequently generate multiple recommended creative image scores and recommended main image scores.
  • adjusting visual preferences on the historical image encoding vector sequence can effectively solve the problem of large amount of calculation.
  • the execution entity may input the historical browsing creative image into a pre-trained graph coding model to generate a historical image coding vector.
  • the graph encoding model may be a model that encodes a creative image to generate an encoding vector.
  • the graph encoding model may be a plurality of serially connected convolutional neural networks.
  • the graph coding model includes: a residual network model and a plurality of fully connected layers.
  • the step of inputting the historical browsing creative image into a pre-trained graph coding model to generate a historical image coding vector may include the following steps:
  • the first step is to input the above historical browsing creative images into the above residual network (ResNets, Residual Networks) model to obtain the model output results.
  • ResNets Residual Networks
  • the second step is to input the output results of the above model into the above multiple fully connected layers to obtain the above
  • the historical image coding vector is described.
  • the multiple fully connected layers can be multiple fully connected layers connected in series.
  • Step 303 Encode each user sparse feature information in the above user sparse feature information set to generate a user sparse feature vector, and obtain a user sparse feature vector set; and encode each item sparse feature information in the above item sparse feature information set to generate an item sparse feature vector, and obtain an item sparse feature vector set.
  • the execution subject may encode each user sparse feature information in the user sparse feature information set to generate a user sparse feature vector, and obtain a user sparse feature vector set, and may encode each item sparse feature information in the item sparse feature information set to generate an item sparse feature vector, and obtain an item sparse feature vector set.
  • the user sparse feature vector may represent feature information of the user sparse feature information.
  • the item sparse feature vector may represent feature information of the item sparse feature information.
  • the execution subject may input each user sparse feature information in the user sparse feature information set into the Bert encoding model to generate a user sparse feature vector and obtain a user sparse feature vector set.
  • the execution subject may input each item sparse feature information in the item sparse feature information set into the Bert encoding model to generate an item sparse feature vector and obtain an item sparse feature vector set.
  • Step 304 performing visual preference adjustment on each historical image coding vector in the above historical image coding vector sequence to obtain an adjusted historical image coding vector sequence.
  • the execution subject may perform visual preference adjustment on each historical image coding vector in the historical image coding vector sequence to obtain an adjusted historical image coding vector sequence.
  • the visual preference information corresponding to each adjusted historical image coding vector included in the adjusted historical image coding vector sequence is the same as the visual preference information corresponding to each historical image coding vector included in the historical image coding vector sequence.
  • the visual preference information may characterize the visual preference characteristics of the user.
  • the video preference characteristics of the user may be, but are not limited to, one of the following: a user's aesthetic video preference characteristics, a user's funny video preference characteristics.
  • the visual preference information corresponding to each adjusted historical image coding vector may be the user's visual preference characteristics reflected by each adjusted historical image coding vector.
  • the visual preference information corresponding to each historical image coding vector may be the visual preference characteristics reflected by each historical image coding vector.
  • the number of historical image coding vectors included in the above historical image coding vector sequence is the same as the number of adjusted historical image coding vectors included in the adjusted historical image coding vector sequence.
  • the above-mentioned execution entity can directly input each historical image coding vector in the historical image coding vector sequence into the Transformer model to generate an adjusted historical image coding vector sequence.
  • each adjusted historical image coding vector included in the adjusted historical image coding vector sequence are stronger than the visual preference characteristics that can be reflected by each historical image coding vector included in the historical image coding vector sequence.
  • the above-mentioned performing visual preference adjustment on each historical image coding vector in the above-mentioned historical image coding vector sequence to obtain the adjusted historical image coding vector sequence may include the following steps:
  • the first step is to determine the visual preference information corresponding to the above historical image coding vector sequence as the target visual preference information.
  • the above-mentioned execution entity can input the above-mentioned historical image encoding vector sequence into a Seq2Seq (Sequence to Sequence) model to output target visual preference information.
  • Seq2Seq Sequence to Sequence
  • each historical image coding vector in the historical image coding vector sequence is adjusted according to the target visual preference information to obtain an adjusted historical image coding vector sequence.
  • the above-mentioned execution entity can input the target visual preference information and each historical image coding vector in the above-mentioned historical image coding vector sequence into a generative adversarial neural network (GAN) model to obtain an adjusted historical image coding vector sequence.
  • GAN generative adversarial neural network
  • Step 305 based on the adjusted historical image coding vector sequence, the main image coding vector, the user sparse feature vector set and the item sparse feature vector set, a multi-head attention mechanism model is used to generate a recommended image set corresponding to the target recommended item to be pushed to the target user.
  • the execution subject may perform the above-mentioned operation according to the above-mentioned adjusted historical image coding vector sequence, the above-mentioned main image coding vector, the above-mentioned user sparse feature vector set and the above-mentioned object
  • the multi-head attention mechanism model generates a recommended image set corresponding to the target recommended item to be pushed to the target user.
  • the recommended image set is an image subset of the creative image set corresponding to the creative image feature information set.
  • the diversified feature information that can be learned by the multi-head attention mechanism model may include but is not limited to at least one of the following: the vector correlation relationship between each adjusted historical image coding vector in the adjusted historical image coding vector sequence, the vector correlation relationship between each user sparse feature vector in the user sparse feature vector set, the vector correlation relationship between each item sparse feature vector in the item sparse feature vector set, the vector correlation relationship between the adjusted historical image coding vector sequence and the main image coding vector, the vector correlation relationship between the adjusted historical image coding vector sequence and the adjusted historical image coding vector sequence, the vector correlation relationship between the adjusted historical image coding vector sequence and the item sparse feature vector set, the vector correlation relationship between the main image coding vector and the user sparse feature vector set, the vector correlation relationship between the main image coding vector and the item sparse feature vector set, and the vector correlation relationship between the user sparse feature vector set and the item sparse feature vector set.
  • the above-mentioned execution entity can input the above-mentioned adjusted historical image encoding vector sequence, the above-mentioned main image encoding vector, the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set into the multi-head attention mechanism model to generate a recommended image set corresponding to the above-mentioned target recommended item to be pushed to the above-mentioned target user.
  • the above-mentioned various embodiments of the present disclosure have the following beneficial effects: through the recommendation information generation method of some embodiments of the present disclosure, accurate recommendation information is generated, and a good recommendation effect is obtained.
  • the reason for the poor recommendation effect is that only the historical browsing creative image sequence of the target user is used to train the user material preference model, so that the feature information that the user material preference model can learn is limited, so that the user material preference model is not accurate enough. This indirectly leads to poor recommendation effect.
  • the recommendation information generation method of some embodiments of the present disclosure first obtains the historical browsing creative image sequence of the target user, the user sparse feature information set of the above target user, the main image of the target recommended item, and the item sparse feature information set of the above target recommended item, wherein the above item sparse feature information set includes the creative image feature information set for the above target recommended item. This is used to subsequently obtain more feature information, so as to facilitate the subsequent generation of more accurate recommendation information (i.e., multiple recommended creative image scores and recommended main image scores).
  • the above main image The image is subjected to image coding processing to obtain the main image coding vector, and each historical browsing creative image in the above-mentioned historical browsing creative image sequence is subjected to image coding processing to generate a historical image coding vector to obtain a historical image coding vector sequence.
  • the main image is subjected to image coding processing to extract the feature information of the main image.
  • generating recommendation information using the main image coding vector can effectively solve the problem of large amount of calculation due to large image pixel dimensions.
  • adjusting visual preference of the historical image coding vector sequence can effectively solve the problem of large amount of calculation.
  • information encoding is performed on each user sparse feature information in the above-mentioned user sparse feature information set to convert it into a vector form, so as to facilitate the use of the feature information of the user sparse feature.
  • information encoding is performed on each item sparse feature information in the above-mentioned item sparse feature information set to convert it into a vector form, so as to facilitate the use of the feature information of the item sparse feature.
  • visual preference adjustment is performed on each historical image coding vector in the above-mentioned historical image coding vector sequence, so that the visual preference reflected by the adjusted historical image coding vector sequence is more obvious, which is helpful for the subsequent generation of more accurate recommendation information.
  • the method for generating recommendation information includes the following steps:
  • Step 401 obtaining a target user's historical browsing creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item.
  • Step 402 performing image coding processing on the main image to obtain a main image coding vector, and performing image coding processing on each historically browsed creative image in the historically browsed creative image sequence to generate a historical image coding vector to obtain a historical image coding vector sequence.
  • Step 403 Encode the sparse feature information of each user in the above user sparse feature information set to generate a user sparse feature vector, and obtain the user sparse feature vector A set is obtained, and information encoding is performed on each item sparse feature information in the item sparse feature information set to generate an item sparse feature vector, thereby obtaining an item sparse feature vector set.
  • Step 404 performing visual preference adjustment on each historical image coding vector in the above historical image coding vector sequence to obtain an adjusted historical image coding vector sequence.
  • steps 401-404 and the technical effects brought about by them can refer to steps 301-304 in the embodiment corresponding to FIG. 3, and will not be repeated here.
  • Step 405 concatenate the user sparse feature vector set and the item sparse feature vector set to obtain a concatenated sparse feature vector.
  • the execution entity may perform vector concatenation on the user sparse feature vector set and the item sparse feature vector set to obtain a concatenated sparse feature vector.
  • Step 406 input the adjusted historical image coding vector sequence, the main image coding vector, and the spliced sparse feature vector into the multi-head attention mechanism model to obtain a feature information fusion vector.
  • the execution subject may input the adjusted historical image coding vector sequence, the main image coding vector, and the spliced sparse feature vector into the multi-head attention mechanism model to obtain a feature information fusion vector.
  • the feature information fusion vector is obtained by fusing multiple feature information.
  • the feature information fusion vector may include but is not limited to at least one of the following: vector relationship information between each historical image coding vector in the adjusted historical image coding vector sequence, vector relationship information between the adjusted historical image coding vector sequence and the main image coding vector, vector relationship information between the spliced sparse feature vector and the adjusted historical image coding vector sequence, and vector relationship information between the spliced sparse feature vector and the main image coding vector.
  • the diversified feature information that can be learned by the multi-head attention mechanism model can also include: the importance information of sparse feature information (i.e., user sparse feature vector sets and item sparse feature vector sets) and the importance information of material content information (i.e., the above-mentioned adjusted historical image encoding vector sequence and the above-mentioned main image encoding vector).
  • the importance information of sparse feature information i.e., user sparse feature vector sets and item sparse feature vector sets
  • the importance information of material content information i.e., the above-mentioned adjusted historical image encoding vector sequence and the above-mentioned main image encoding vector.
  • Step 407 input the concatenated sparse feature vector into a fully connected model to obtain an output vector.
  • Step 408 using a preset loss function to generate the recommended image set.
  • the execution subject may use a preset loss function to generate the recommended image set in various ways.
  • the preset loss function may be a quadratic loss function.
  • the above-mentioned generation of the above-mentioned recommended image set by using a preset loss function may include the following steps:
  • the preset loss function is used to generate a creative image score set for the splicing vector.
  • the creative image score represents the interest of the target user in the creative image of the target recommended item.
  • the creative image score set in the creative image score set corresponds to the creative image in the creative image set.
  • the execution entity may input the stitching vector into a preset loss function to obtain a creative image score set for the stitching vector.
  • the second step is to generate the recommended item score corresponding to the target user using the multi-head attention mechanism model based on the adjusted historical image coding vector sequence, the main image coding vector, the user sparse feature vector set and the item sparse feature vector set.
  • the recommended item score can represent the target user's preference for the target recommended item.
  • the above-mentioned execution entity can input the above-mentioned adjusted historical image encoding vector sequence, the above-mentioned main image encoding vector, the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set into the multi-head attention mechanism model to generate the recommended item score corresponding to the above-mentioned target user.
  • the execution entity may determine the creative image scores in the creative image score set whose score values are greater than the predetermined recommended creative image value, and obtain a creative image score subset.
  • the value of the predetermined recommended item is 75 points.
  • the execution entity may determine the creative image set corresponding to the creative image score subset as the recommended image set.
  • the steps further include:
  • the first step is to use the above adjusted historical image coding vector sequence and the above main image coding vector sequence.
  • the vector, the user sparse feature vector set and the item sparse feature vector set are used to generate a recommended main image score corresponding to the target recommended item using a multi-head attention mechanism model.
  • the recommended main image score represents the interest of the target user in the main image.
  • the above-mentioned execution entity can input the above-mentioned adjusted historical image encoding vector sequence, the above-mentioned main image encoding vector, the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set into the multi-head attention mechanism model to generate the recommended main image score corresponding to the above-mentioned target recommended item.
  • the main image is pushed to a terminal corresponding to the target user.
  • the terminal corresponding to the target user may be a display terminal.
  • the predetermined recommended item value may be 70.
  • the predetermined recommended main image value may be 75.
  • the process 400 of the recommendation information generation method in some embodiments corresponding to Figure 4 using a multi-head attention mechanism model, can learn the diversified feature information of the importance information of sparse feature information (i.e., user sparse feature vector set and item sparse feature vector set) and the importance information of material content information (i.e., the above-mentioned adjusted historical image coding vector sequence and the above-mentioned main image coding vector), so as to generate a more accurate creative image set.
  • sparse feature information i.e., user sparse feature vector set and item sparse feature vector set
  • material content information i.e., the above-mentioned adjusted historical image coding vector sequence and the above-mentioned main image coding vector
  • the present disclosure provides some embodiments of a device for generating recommendation information. These device embodiments correspond to the method embodiments shown in FIG. 3 , and the device can be specifically applied to various electronic devices.
  • a recommendation information generating device 500 includes: an acquisition unit 501, a graph encoding unit 502, an information encoding unit 503, an adjustment unit 504, and a generation unit 505.
  • the acquisition unit 501 is configured to acquire a target user's historical browsing creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item, wherein the item sparse feature information set includes a creative image feature information set for the target recommended item;
  • the graph encoding unit 502 is configured to acquire a target user's historical browsing creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item;
  • Element 502 is configured to perform image coding processing on the above-mentioned main image to obtain a main image coding vector, and to perform image coding processing on each historical browsing creative image in the above-mentioned historical
  • the adjustment unit 504 in the above-mentioned device 500 can be configured to: determine the visual preference information corresponding to the above-mentioned historical image coding vector sequence as the target visual preference information; and adjust each historical image coding vector in the above-mentioned historical image coding vector sequence according to the above-mentioned target visual preference information to obtain an adjusted historical image coding vector sequence.
  • the generation unit 505 in the above-mentioned device 500 can be configured to: perform vector splicing on the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set to obtain a spliced sparse feature vector; input the above-mentioned adjusted historical image encoding vector sequence, the above-mentioned main image encoding vector, and the above-mentioned spliced sparse feature vector into the above-mentioned multi-head attention mechanism model to obtain a feature information fusion vector; input the above-mentioned spliced sparse feature vector into the fully connected model to obtain an output vector; splice the above-mentioned feature information fusion vector and the above-mentioned output vector to obtain a spliced vector; and use a preset loss function to generate the above-mentioned recommended image set.
  • the graph coding model includes: a residual network model and a plurality of fully connected layers; the graph coding unit 502 in the apparatus 500 may be configured to: input the historical browsing creative image into the residual network model to obtain a model output result; input the model output result into the plurality of fully connected layers to obtain to the above historical image encoding vector.
  • the generation unit 505 in the above-mentioned device 500 can be configured to: generate a creative image score set for the above-mentioned splicing vector using the above-mentioned preset loss function; generate a recommended item score corresponding to the above-mentioned target recommended item using a multi-head attention mechanism model based on the above-mentioned adjusted historical image coding vector sequence, the above-mentioned main image coding vector, the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set; in response to determining that the above-mentioned recommended item score is greater than a predetermined recommended item value, determine the creative image score in the above-mentioned creative image score set whose score value is greater than a predetermined recommended creative image value, and obtain a creative image score subset; determine the creative image set corresponding to the above-mentioned creative image score subset as the above-mentioned recommended image set.
  • the generation unit 505 in the above-mentioned device 500 can be configured to: generate a recommended main image score corresponding to the above-mentioned target recommended item based on the above-mentioned adjusted historical image encoding vector sequence, the above-mentioned main image encoding vector, the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set using a multi-head attention mechanism model; in response to determining that the above-mentioned recommended item score is greater than a predetermined recommended item value and the above-mentioned recommended main image score is greater than a predetermined recommended main image value, push the above-mentioned main image to the terminal corresponding to the above-mentioned target user.
  • FIG6 a schematic diagram of an electronic device (such as the electronic device 101 in FIG1) 600 suitable for implementing some embodiments of the present disclosure is shown.
  • the electronic device shown in FIG6 is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure.
  • the electronic device 600 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which may perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 to a random access memory (RAM) 603.
  • the RAM 603 also stores the electronic device Various programs and data required for the operation of the bus 600.
  • the processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604.
  • An input/output (I/O) interface 605 is also connected to the bus 604.
  • the following devices may be connected to the I/O interface 605: input devices 606 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 608 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 609.
  • the communication device 609 may allow the electronic device 600 to communicate wirelessly or wired with other devices to exchange data.
  • FIG. 6 shows an electronic device 600 with various devices, it should be understood that it is not required to implement or have all the devices shown. More or fewer devices may be implemented or have alternatively. Each box shown in FIG. 6 may represent one device, or may represent multiple devices as needed.
  • the process described above with reference to the flowchart can be implemented as a computer software program.
  • some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart.
  • the computer program can be downloaded and installed from the network through the communication device 609, or installed from the storage device 608, or installed from the ROM 602.
  • the processing device 601 the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.
  • the above-mentioned computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two.
  • the computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above.
  • Computer-readable storage media may include, but are not limited to: an electrical connection with at least one wire, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.
  • the computer-readable storage medium may be any tangible medium that contains or stores a program, The program can be used by or in combination with an instruction execution system, device or device.
  • a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code.
  • This propagated data signal can take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above.
  • the computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate or transmit a program for use by or in combination with an instruction execution system, device or device.
  • the program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.
  • the client and server may communicate using any currently known or future developed network protocol such as HTTP (Hyper Text Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network).
  • HTTP Hyper Text Transfer Protocol
  • Examples of communication networks include a local area network ("LAN”), a wide area network ("WAN”), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.
  • the computer-readable medium may be included in the electronic device; or it may exist independently without being installed in the electronic device.
  • the computer-readable medium carries one or more programs.
  • the electronic device obtains the target user's historical browsing creative image sequence, the target user's user sparse feature information set, the target recommended item's main image, and the target recommended item's item sparse feature information set, wherein the item sparse feature information set includes a creative image feature information set for the target recommended item; performs image coding processing on the main image to obtain a main image coding vector, and performs image coding processing on each historical browsing creative image in the historical browsing creative image sequence to generate a historical image coding vector to obtain a historical image coding vector sequence; performs information coding on each user sparse feature information in the user sparse feature information set to generate a user sparse feature vector to obtain a user sparse feature vector set, and performs information coding on each item sparse feature information in the item sparse feature information
  • Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages.
  • the program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server.
  • the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
  • LAN local area network
  • WAN wide area network
  • Internet service provider e.g., via the Internet using an Internet service provider
  • each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains at least one executable instruction for realizing the specified logical function.
  • the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved.
  • each square box in the block diagram and/or flow chart, and the combination of the square boxes in the block diagram and/or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
  • the units described in some embodiments of the present disclosure may be implemented by software or hardware.
  • the units described may also be set in a processor.
  • a processor includes an acquisition unit, a graph encoding unit, an information encoding unit, an adjustment unit, and a generation unit.
  • the names of these units do not, in some cases, limit the units themselves.
  • the acquisition unit may also be described as “acquisition unit”. Take a unit of a target user's historical browsing creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item.
  • exemplary types of hardware logic components include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.
  • FPGAs field programmable gate arrays
  • ASICs application specific integrated circuits
  • ASSPs application specific standard products
  • SOCs systems on chip
  • CPLDs complex programmable logic devices
  • Some embodiments of the present disclosure further provide a computer program product, including a computer program, which implements any of the above-mentioned recommendation information generation methods when executed by a processor.

Abstract

Disclosed in the embodiments of the present disclosure are a recommendation-information generation method and apparatus, and a device, a medium and a program product. A specific embodiment of the method comprises: acquiring a sequence of historically browsed creative images, a user sparse-feature information set, a main image and an article sparse-feature information set; performing image coding processing on the main image to obtain a main-image coding vector, and performing image coding processing on each historically browsed creative image to generate a historical image coding vector; performing information coding on each piece of user sparse-feature information to generate a user sparse-feature vector, and performing information coding on each piece of article sparse-feature information to generate an article sparse-feature vector; performing visual preference adjustment on each historical image coding vector, so as to obtain an adjusted historical image coding vector sequence; and generating a recommended image set which corresponds to a target recommended article and is to be pushed to a target user.

Description

推荐信息生成方法、装置、设备、介质和程序产品Recommendation information generation method, device, equipment, medium and program product
相关申请的交叉引用CROSS-REFERENCE TO RELATED APPLICATIONS
本申请要求于申请日为2022年09月28日提交的，申请号为202211194196.3、发明名称为“推荐信息生成方法、装置、设备、介质和程序产品”的中国专利申请的优先权，其全部内容作为整体并入本申请中。This application claims priority to the Chinese patent application filed on September 28, 2022, with application number 202211194196.3 and invention name “Recommendation information generation method, device, equipment, medium and program product”, all contents of which are incorporated into this application as a whole.
技术领域Technical Field
本公开的实施例涉及计算机技术领域，具体涉及推荐信息生成方法、装置、设备、介质和程序产品。Embodiments of the present disclosure relate to the field of computer technology, and in particular to a method, apparatus, device, medium, and program product for generating recommendation information.
背景技术Background technique
当前，物品往往存在多个预先设计的素材(例如，物品主图，创意图像等等)。对于从多个素材中筛选出目标用户可能喜好的素材，通常采用的方式为：首先，将多个素材输入至预先训练的用户素材偏好模型，以生成多个素材对应的分数集。然后，利用分数集，从上述多个素材中筛选出目标用户可能喜好的至少一个素材。最后，将至少一个素材推荐给目标用户。Currently, there are often multiple pre-designed materials for items (e.g., main images of items, creative images, etc.). To filter out materials that the target user may like from the multiple materials, the usual method is: first, input the multiple materials into a pre-trained user material preference model to generate a score set corresponding to the multiple materials. Then, use the score set to filter out at least one material that the target user may like from the multiple materials. Finally, recommend the at least one material to the target user.
发明内容Summary of the invention
本公开的内容部分用于以简要的形式介绍构思，这些构思将在后面的具体实施方式部分被详细描述。本公开的内容部分并不旨在标识要求保护的技术方案的关键特征或必要特征，也不旨在用于限制所要求的保护的技术方案的范围。The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.
本公开的一些实施例提出了推荐信息生成方法、装置、电子设备、计算机可读介质和计算机程序产品，来解决以上背景技术部分提到的技术问题。 Some embodiments of the present disclosure propose a recommendation information generation method, an apparatus, an electronic device, a computer-readable medium, and a computer program product to solve the technical problems mentioned in the above background technology section.
第一方面，本公开的一些实施例提供了一种推荐信息生成方法，包括：获取目标用户的历史浏览创意图像序列、上述目标用户的用户稀疏特征信息集、目标推荐物品的主图像和上述目标推荐物品的物品稀疏特征信息集，其中，上述物品稀疏特征信息集包括针对上述目标推荐物品的创意图像特征信息集；对上述主图像进行图编码处理，得到主图编码向量，以及对上述历史浏览创意图像序列中的每个历史浏览创意图像进行图编码处理，以生成历史图像编码向量，得到历史图像编码向量序列；对上述用户稀疏特征信息集中的每个用户稀疏特征信息进行信息编码，以生成用户稀疏特征向量，得到用户稀疏特征向量集，以及对上述物品稀疏特征信息集中的每个物品稀疏特征信息进行信息编码，以生成物品稀疏特征向量，得到物品稀疏特征向量集；对上述历史图像编码向量序列中的各个历史图像编码向量进行视觉偏好调整，得到调整后历史图像编码向量序列；根据上述调整后历史图像编码向量序列、上述主图编码向量、上述用户稀疏特征向量集和上述物品稀疏特征向量集，利用多头注意力机制模型，生成待推送给上述目标用户的、上述目标推荐物品对应的推荐图像集。In a first aspect, some embodiments of the present disclosure provide a method for generating recommendation information, comprising: obtaining a target user's historically browsed creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item, wherein the item sparse feature information set includes a creative image feature information set for the target recommended item; performing graph coding processing on the main image to obtain a main image coding vector, and performing graph coding processing on each historically browsed creative image in the historically browsed creative image sequence to generate a historical image coding vector to obtain a historical image coding vector sequence; performing graph coding processing on each user sparse feature information set The feature information is encoded to generate a user sparse feature vector to obtain a user sparse feature vector set, and the sparse feature information of each item in the above-mentioned item sparse feature information set is encoded to generate an item sparse feature vector to obtain an item sparse feature vector set; the visual preference of each historical image coding vector in the above-mentioned historical image coding vector sequence is adjusted to obtain an adjusted historical image coding vector sequence; according to the above-mentioned adjusted historical image coding vector sequence, the above-mentioned main image coding vector, the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set, a multi-head attention mechanism model is used to generate a recommended image set corresponding to the above-mentioned target recommended item to be pushed to the above-mentioned target user.
第二方面，本公开的一些实施例提供了一种推荐信息生成装置，包括：获取单元，被配置成获取目标用户的历史浏览创意图像序列、上述目标用户的用户稀疏特征信息集、目标推荐物品的主图像和上述目标推荐物品的物品稀疏特征信息集，其中，上述物品稀疏特征信息集包括针对上述目标推荐物品的创意图像特征信息集；图编码单元，被配置成对上述主图像进行图编码处理，得到主图编码向量，以及对上述历史浏览创意图像序列中的每个历史浏览创意图像进行图编码处理，以生成历史图像编码向量，得到历史图像编码向量序列；信息编码单元，被配置成对上述用户稀疏特征信息集中的每个用户稀疏特征信息进行信息编码，以生成用户稀疏特征向量，得到用户稀疏特征向量集，以及对上述物品稀疏特征信息集中的每个物品稀疏特征信息进行信息编码，以生成物品稀疏特征向量，得到物品稀疏特征向量集；调整单元，被配置成对上述历史图像编码向量序列中的各个历史图像编码向量进行视觉偏好调整，得到调整后历史图像编码向量序列；生 成单元，被配置成根据上述调整后历史图像编码向量序列、上述主图编码向量、上述用户稀疏特征向量集和上述物品稀疏特征向量集，利用多头注意力机制模型，生成待推送给上述目标用户的、上述目标推荐物品对应的推荐图像集。In a second aspect, some embodiments of the present disclosure provide a recommendation information generating device, including: an acquisition unit, configured to acquire a target user's historical browsing creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item, wherein the item sparse feature information set includes a creative image feature information set for the target recommended item; a graph encoding unit, configured to perform graph encoding processing on the main image to obtain a main graph encoding vector, and to perform graph encoding processing on each historical browsing creative image in the historical browsing creative image sequence to generate a historical image encoding vector to obtain a historical image encoding vector sequence; an information encoding unit, configured to perform information encoding on each user sparse feature information in the user sparse feature information set to generate a user sparse feature vector to obtain a user sparse feature vector set, and to perform information encoding on each item sparse feature information in the item sparse feature information set to generate an item sparse feature vector to obtain an item sparse feature vector set; an adjustment unit, configured to perform visual preference adjustment on each historical image encoding vector in the historical image encoding vector sequence to obtain an adjusted historical image encoding vector sequence; The unit is configured to generate a recommended image set corresponding to the target recommended item to be pushed to the target user by using a multi-head attention mechanism model based on the adjusted historical image coding vector sequence, the main image coding vector, the user sparse feature vector set and the item sparse feature vector set.
第三方面，本公开的一些实施例提供了一种电子设备，包括：至少一个处理器；存储装置，其上存储有至少一个程序，当至少一个程序被至少一个处理器执行，使得至少一个处理器实现如第一方面中任一实现方式描述的方法。In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: at least one processor; and a storage device on which at least one program is stored, wherein when the at least one program is executed by the at least one processor, the at least one processor implements the method described in any implementation manner in the first aspect.
第四方面，本公开的一些实施例提供了一种计算机可读介质，其上存储有计算机程序，其中，程序被处理器执行时实现如第一方面中任一实现方式描述的方法。In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation manner in the first aspect is implemented.
第五方面，本公开的一些实施例提供了一种计算机程序产品，包括计算机程序，计算机程序在被处理器执行时实现上述第一方面中任一实现方式所描述的方法。In a fifth aspect, some embodiments of the present disclosure provide a computer program product, including a computer program, which implements the method described in any implementation manner in the above-mentioned first aspect when executed by a processor.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
结合附图并参考以下具体实施方式，本公开各实施例的上述和其他特征、优点及方面将变得更加明显。贯穿附图中，相同或相似的附图标记表示相同或相似的元素。应当理解附图是示意性的，元件和元素不一定按照比例绘制。The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.
图1-图2是根据本公开的一些实施例的推荐信息生成方法的一个应用场景的示意图；1-2 are schematic diagrams of an application scenario of a method for generating recommendation information according to some embodiments of the present disclosure;
图3是根据本公开的推荐信息生成方法的一些实施例的流程图；FIG3 is a flow chart of some embodiments of a method for generating recommendation information according to the present disclosure;
图4是根据本公开的推荐信息生成方法的另一些实施例的流程图；FIG4 is a flow chart of other embodiments of the method for generating recommendation information according to the present disclosure;
图5是根据本公开的推荐信息生成装置的一些实施例的结构示意图；FIG5 is a schematic diagram of the structure of some embodiments of a device for generating recommendation information according to the present disclosure;
图6是适于用来实现本公开的一些实施例的电子设备的结构示意图。 FIG. 6 is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure.
具体实施方式Detailed ways
下面将参照附图更详细地描述本公开的实施例。虽然附图中显示了本公开的某些实施例，然而应当理解的是，本公开可以通过各种形式来实现，而且不应该被解释为限于这里阐述的实施例。相反，提供这些实施例是为了更加透彻和完整地理解本公开。应当理解的是，本公开的附图及实施例仅用于示例性作用，并非用于限制本公开的保护范围。Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.
另外还需要说明的是，为了便于描述，附图中仅示出了与有关发明相关的部分。在不冲突的情况下，本公开中的实施例及实施例中的特征可以相互组合。It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other.
需要注意，本公开中提及的“第一”、“第二”等概念仅用于对不同的装置、模块或单元进行区分，并非用于限定这些装置、模块或单元所执行的功能的顺序或者相互依存关系。It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.
需要注意，本公开中提及的“一个”、“多个”的修饰是示意性而非限制性的，本领域技术人员应当理解，除非在上下文另有明确指出，否则应该理解为“至少一个”。It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "at least one".
本公开实施方式中的多个装置之间所交互的消息或者信息的名称仅用于说明性的目的，而并不是用于对这些消息或信息的范围进行限制。The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.
本公开中所涉及的创意素材(例如创意图像、主图像)的收集、存储、使用等操作，在执行相应操作之前，相关组织或个人尽到包括开展素材安全影响评估、向素材创造主体履行告知义务、事先征得素材创造主体的授权同意等义务。With regard to the collection, storage, and use of creative materials (such as creative images and main images) involved in this disclosure, before executing corresponding operations, relevant organizations or individuals shall fulfill obligations including conducting material security impact assessments, fulfilling the obligation to inform the material creators, and obtaining the authorization and consent of the material creators in advance.
相关的推荐信息生成方法，例如，首先，将多个素材输入至预先训练的用户素材偏好模型，以生成多个素材对应的分数集。然后，利用分数集，从上述多个素材中筛选出目标用户可能喜好的至少一个素材。最后，将至少一个素材推荐给目标用户等经常会存在如下技术问题：仅利用针对目标用户的历史浏览创意图像序列，对用户素材偏好模型进行训练，使得用户素材偏好模型所能学习到的特征信息有限，以致用户素材偏好模型不够精准。侧面导致推荐效果不佳。 Related recommendation information generation methods, for example, first, input multiple materials into a pre-trained user material preference model to generate a score set corresponding to the multiple materials. Then, use the score set to filter out at least one material that the target user may like from the above multiple materials. Finally, recommending at least one material to the target user often has the following technical problems: only using the creative image sequence of the target user's historical browsing to train the user material preference model, so that the user material preference model can learn limited feature information, so that the user material preference model is not accurate enough. This leads to poor recommendation results.
为了解决以上所阐述的问题，本公开的一些实施例提出了推荐信息生成方法及装置，生成了精准的推荐信息，得到了较好的推荐效果。In order to solve the above-mentioned problems, some embodiments of the present disclosure propose a method and device for generating recommendation information, which generates accurate recommendation information and achieves better recommendation effect.
下面将参考附图并结合实施例来详细说明本公开。The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.
图1-图2是根据本公开一些实施例的推荐信息生成方法的一个应用场景的示意图。1-2 are schematic diagrams of an application scenario of a method for generating recommendation information according to some embodiments of the present disclosure.
在图1-图2的应用场景中，首先，电子设备101可以获取目标用户102的历史浏览创意图像序列103、上述目标用户102的用户稀疏特征信息集104、目标推荐物品105的主图像106和上述目标推荐物品105的物品稀疏特征信息集107。其中，上述物品稀疏特征信息集107包括针对上述目标推荐物品105的创意图像特征信息集。在本应用场景中，目标用户102可以是“李**”。历史浏览创意图像序列103可以包括：历史浏览创意图像1031和历史浏览创意图像1032。用户稀疏特征信息集104可以是：{“性别：1”，“年龄：18”，“身高：184”}。目标推荐物品105可以是“苹果”。物品稀疏特征信息集107可以是：{“点击率：0.4”，“价格：5”，“产地：02”}。其中，性别“1”可以表征男性。产地“02”可以表征上海。然后，电子设备101可以对上述主图像106进行图编码处理，得到主图编码向量110，以及对上述历史浏览创意图像序列103中的每个历史浏览创意图像进行图编码处理，以生成历史图像编码向量，得到历史图像编码向量序列108。在本应用场景中，历史图像编码向量序列108可以包括：历史浏览创意图像1031对应的历史图像编码向量1081和历史浏览创意图像1032对应的历史图像编码向量1082。接着，电子设备101可以对上述用户稀疏特征信息集104中的每个用户稀疏特征信息进行信息编码，以生成用户稀疏特征向量，得到用户稀疏特征向量集109，以及对上述物品稀疏特征信息集107中的每个物品稀疏特征信息进行信息编码，以生成物品稀疏特征向量，得到物品稀疏特征向量集111。在本应用场景中，用户稀疏特征向量集109可以包括：“性别：1”对应的用户稀疏特征向量1091、“年龄：18”对应的用户稀疏特征向量1092和“身高：184”对应的用户稀疏特征向量1093。物品稀疏特征向量集111可以包括：“点击率：0.4”对应的物品稀疏特征向量1111、“价格：5” 对应的物品稀疏特征向量1112和“产地：02”对应的物品稀疏特征向量1113。进而，电子设备101可以对上述历史图像编码向量序列108中的各个历史图像编码向量进行视觉偏好调整，得到调整后历史图像编码向量序列112。在本应用场景中，调整后历史图像编码向量序列112包括：历史图像编码向量1081对应的调整后历史图像编码向量1121和历史图像编码向量1082对应的调整后历史图像编码向量1122。最后，电子设备101可以根据上述调整后历史图像编码向量序列112、上述主图编码向量110、上述用户稀疏特征向量集109和上述物品稀疏特征向量集111，利用多头注意力机制模型113，生成待推送给上述目标用户102的、上述目标推荐物品105对应的推荐图像集114。其中，上述推荐图像集114为上述创意图像特征信息集对应创意图像集的图像子集。在本应用场景中，推荐图像集114包括：推荐图像1141和推荐图像1142。In the application scenarios of FIG. 1 and FIG. 2, first, the electronic device 101 can obtain the historical browsing creative image sequence 103 of the target user 102, the user sparse feature information set 104 of the target user 102, the main image 106 of the target recommended item 105, and the item sparse feature information set 107 of the target recommended item 105. Among them, the item sparse feature information set 107 includes a creative image feature information set for the target recommended item 105. In this application scenario, the target user 102 can be "Li**". The historical browsing creative image sequence 103 can include: historical browsing creative images 1031 and historical browsing creative images 1032. The user sparse feature information set 104 can be: {"gender: 1", "age: 18", "height: 184"}. The target recommended item 105 can be "apple". The item sparse feature information set 107 can be: {"click rate: 0.4", "price: 5", "origin: 02"}. Among them, the gender "1" can represent male. The origin "02" can represent Shanghai. Then, the electronic device 101 may perform image coding processing on the main image 106 to obtain a main image coding vector 110, and perform image coding processing on each historical browsing creative image in the historical browsing creative image sequence 103 to generate a historical image coding vector to obtain a historical image coding vector sequence 108. In this application scenario, the historical image coding vector sequence 108 may include: a historical image coding vector 1081 corresponding to the historical browsing creative image 1031 and a historical image coding vector 1082 corresponding to the historical browsing creative image 1032. Next, the electronic device 101 may perform information coding on each user sparse feature information in the user sparse feature information set 104 to generate a user sparse feature vector to obtain a user sparse feature vector set 109, and perform information coding on each item sparse feature information in the item sparse feature information set 107 to generate an item sparse feature vector to obtain an item sparse feature vector set 111. In this application scenario, the user sparse feature vector set 109 may include a user sparse feature vector 1091 corresponding to "gender: 1", a user sparse feature vector 1092 corresponding to "age: 18", and a user sparse feature vector 1093 corresponding to "height: 184". The item sparse feature vector set 111 may include an item sparse feature vector 1111 corresponding to "click rate: 0.4", and an item sparse feature vector 1112 corresponding to "price: 5". The corresponding item sparse feature vector 1112 and the item sparse feature vector 1113 corresponding to "Origin: 02". Furthermore, the electronic device 101 can adjust the visual preference of each historical image coding vector in the above historical image coding vector sequence 108 to obtain an adjusted historical image coding vector sequence 112. In this application scenario, the adjusted historical image coding vector sequence 112 includes: an adjusted historical image coding vector 1121 corresponding to the historical image coding vector 1081 and an adjusted historical image coding vector 1122 corresponding to the historical image coding vector 1082. Finally, the electronic device 101 can generate a recommended image set 114 corresponding to the above target recommended item 105 to be pushed to the above target user 102 using the multi-head attention mechanism model 113 based on the above adjusted historical image coding vector sequence 112, the above main image coding vector 110, the above user sparse feature vector set 109 and the above item sparse feature vector set 111. Among them, the above recommended image set 114 is an image subset of the creative image set corresponding to the above creative image feature information set. In this application scenario, the recommended image set 114 includes: a recommended image 1141 and a recommended image 1142 .
需要说明的是，上述电子设备101可以是硬件，也可以是软件。当电子设备为硬件时，可以实现成多个服务器或终端设备组成的分布式集群，也可以实现成单个服务器或单个终端设备。当电子设备体现为软件时，可以安装在上述所列举的硬件设备中。其可以实现成例如用来提供分布式服务的多个软件或软件模块，也可以实现成单个软件或软件模块。在此不做具体限定。It should be noted that the electronic device 101 can be hardware or software. When the electronic device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or it can be implemented as a single server or a single terminal device. When the electronic device is embodied as software, it can be installed in the hardware devices listed above. It can be implemented as multiple software or software modules for providing distributed services, for example, or it can be implemented as a single software or software module. No specific limitation is made here.
应该理解，图1-图2中的电子设备的数目仅仅是示意性的。根据实现需要，可以具有任意数目的电子设备。It should be understood that the number of electronic devices in FIG. 1-FIG 2 is only for illustration purposes, and any number of electronic devices may be provided according to implementation requirements.
继续参考图3，示出了根据本公开的推荐信息生成方法的一些实施例的流程300。该推荐信息生成方法，包括以下步骤：Continuing to refer to FIG3 , a process 300 of some embodiments of the method for generating recommendation information according to the present disclosure is shown. The method for generating recommendation information includes the following steps:
步骤301，获取目标用户的历史浏览创意图像序列、上述目标用户的用户稀疏特征信息集、目标推荐物品的主图像和上述目标推荐物品的物品稀疏特征信息集。Step 301, obtaining a target user's historical browsing creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item.
在一些实施例中，上述推荐信息生成方法的执行主体(例如图1所示的电子设备101)可以通过有线连接方式或者无线连接方式来获取目标用户的历史浏览创意图像序列、上述目标用户的用户稀疏特征 信息集、目标推荐物品的主图像和上述目标推荐物品的物品稀疏特征信息集。其中，上述物品稀疏特征信息集包括针对上述目标推荐物品的创意图像特征信息集。创意图像特征信息可以是目标推荐物品的创意图像的ID标识信息。ID标识信息与创意图像存在一一对应关系。其中，目标用户可以是待推荐素材的用户。实践中，素材的表现形式可以是图片的形式，也可以是视频的形式。针对素材为视频的形式，可以将视频中抽取的关键帧作为推荐给用户的内容。针对电商场景，素材可以是物品的主图，也可以是物品的创意图像。创意图像可以是针对物品的特征，预先设计的图像。历史浏览创意图像序列可以是目标用户在历史时间段内在各个应用(Application，App)上所浏览的创意图像序列。例如，当前时间为2022年4月份。历史浏览创意图像序列可以是2021年1月份-2021年4月份的用户浏览创意图像序列。目标推荐物品的主图像可以是目标推荐物品的主图。目标用户的用户稀疏特征信息可以是目标用户的稀疏特征的数值信息。目标用户的稀疏特征可以是目标用户的ID类特征。上述ID类特征可以用户特征的唯一ID标识，也可以是用户身份的标识。例如，目标用户的稀疏特征可以是以下之一：目标用户的性别ID标识，目标用户的年龄ID标识，目标用户的身份证ID标识，目标用户的身高ID标识。再例如，性别ID标识为“1”，表征目标用户为男性。性别ID标识为“0”，表征目标用户为女性。目标推荐物品的物品稀疏特征信息可以是目标推荐物品的稀疏特征的数值信息。目标推荐物品的稀疏特征可以是目标推荐物品的ID类特征。ID类特征可以物品特征的唯一ID标识，也可以是物品身份的标识。例如，目标推荐物品的稀疏特征可以是以下之一：目标推荐物品的点击率，目标推荐物品的价格，目标推荐物品的产地。再例如，产地ID标识为“02”，表征目标推荐物品的产地为上海。In some embodiments, the execution subject of the above-mentioned recommendation information generation method (for example, the electronic device 101 shown in FIG. 1 ) can obtain the target user's historical browsing creative image sequence and the user sparse features of the above-mentioned target user through a wired connection or a wireless connection. Information set, the main image of the target recommended item and the item sparse feature information set of the above target recommended item. Among them, the above item sparse feature information set includes a creative image feature information set for the above target recommended item. The creative image feature information can be the ID identification information of the creative image of the target recommended item. There is a one-to-one correspondence between the ID identification information and the creative image. Among them, the target user can be the user of the material to be recommended. In practice, the material can be in the form of a picture or a video. For the material in the form of a video, the key frames extracted from the video can be used as content recommended to the user. For e-commerce scenarios, the material can be the main picture of the item or the creative image of the item. The creative image can be a pre-designed image based on the characteristics of the item. The historical browsing creative image sequence can be the creative image sequence browsed by the target user on each application (Application, App) in the historical time period. For example, the current time is April 2022. The historical browsing creative image sequence can be the creative image sequence browsed by users from January 2021 to April 2021. The main image of the target recommended item can be the main image of the target recommended item. The user sparse feature information of the target user can be the numerical information of the sparse features of the target user. The sparse features of the target user may be the ID features of the target user. The above-mentioned ID features may be the unique ID identifier of the user features, or the identifier of the user identity. For example, the sparse features of the target user may be one of the following: the gender ID identifier of the target user, the age ID identifier of the target user, the ID identifier of the target user, and the height ID identifier of the target user. For another example, the gender ID identifier is "1", which indicates that the target user is male. The gender ID identifier is "0", which indicates that the target user is female. The sparse feature information of the target recommended item may be the numerical information of the sparse features of the target recommended item. The sparse features of the target recommended item may be the ID features of the target recommended item. The ID features may be the unique ID identifier of the item features, or the identifier of the item identity. For example, the sparse features of the target recommended item may be one of the following: the click-through rate of the target recommended item, the price of the target recommended item, and the origin of the target recommended item. For another example, the origin ID identifier is "02", which indicates that the origin of the target recommended item is Shanghai.
需要说明的是，可以推荐给目标用户的信息可以包括但不限于以下之一：目标推荐物品的创意图像，目标推荐物品的主图。It should be noted that the information that can be recommended to the target user may include but is not limited to one of the following: a creative image of the target recommended item, and a main image of the target recommended item.
步骤302，对上述主图像进行图编码处理，得到主图编码向量，以及对上述历史浏览创意图像序列中的每个历史浏览创意图像进行图编码处理，以生成历史图像编码向量，得到历史图像编码向量序列。 Step 302, performing image coding processing on the main image to obtain a main image coding vector, and performing image coding processing on each historically browsed creative image in the historically browsed creative image sequence to generate a historical image coding vector to obtain a historical image coding vector sequence.
在一些实施例中，上述执行主体可以对上述主图像进行图编码处理，得到主图编码向量，以及对上述历史浏览创意图像序列中的每个历史浏览创意图像进行图编码处理，以生成历史图像编码向量，得到历史图像编码向量序列。主图编码向量可以表征主图像的图像特征信息。历史图像编码向量可以表征历史浏览创意图像的图像特征信息。In some embodiments, the execution subject may perform image coding processing on the main image to obtain a main image coding vector, and perform image coding processing on each historically browsed creative image in the historically browsed creative image sequence to generate a historical image coding vector to obtain a historical image coding vector sequence. The main image coding vector may represent image feature information of the main image. The historical image coding vector may represent image feature information of the historically browsed creative images.
作为示例，首先，上述执行主体可以将主图像输入至多个串行连接的卷积神经网络(Convolutional Neural Networks，CNN)中，得到第一模型输出结果。然后，将第一模型输出结果输入至Bert编码模型，得到主图编码向量。As an example, first, the execution subject may input the main image into a plurality of serially connected convolutional neural networks (CNNs) to obtain a first model output result. Then, the first model output result is input into a Bert coding model to obtain a main image coding vector.
同样地，首先，上述执行主体可以将历史浏览创意图像序列中的每个历史浏览创意图像输入至多个串行连接的卷积神经网络中，以生成第二模型输出结果，得到第二模型输出结果序列。然后，将第二模型输出结果序列中的每个第二模型输出结果输入至Bert编码模型，得到历史图像编码向量序列。Similarly, first, the execution subject can input each historical browsing creative image in the historical browsing creative image sequence into a plurality of serially connected convolutional neural networks to generate a second model output result and obtain a second model output result sequence. Then, each second model output result in the second model output result sequence is input into the Bert coding model to obtain a historical image coding vector sequence.
在这里，利用主图编码向量来后续生成多个推荐创意图像分数和推荐主图像分数，与利用主图像来后续来生成多个推荐创意图像分数和推荐主图像分数来比，计算量更小。相对于通过历史浏览创意图像序列来进行视觉偏好调整，对历史图像编码向量序列进行视觉偏好调整可以有效解决计算量较大的问题。Here, using the main image encoding vector to subsequently generate multiple recommended creative image scores and recommended main image scores has a smaller amount of calculation than using the main image to subsequently generate multiple recommended creative image scores and recommended main image scores. Compared with adjusting visual preferences by browsing the creative image sequence historically, adjusting visual preferences on the historical image encoding vector sequence can effectively solve the problem of large amount of calculation.
在一些实施例的一些可选的实现方式中，上述执行主体可以将上述历史浏览创意图像输入至预先训练的图编码模型，以生成历史图像编码向量。In some optional implementations of some embodiments, the execution entity may input the historical browsing creative image into a pre-trained graph coding model to generate a historical image coding vector.
其中，图编码模型可以是对创意图像进行编码以生成编码向量的模型。例如，图编码模型可以是多个串行连接的卷积神经网络。The graph encoding model may be a model that encodes a creative image to generate an encoding vector. For example, the graph encoding model may be a plurality of serially connected convolutional neural networks.
可选地，上述图编码模型包括：残差网络模型和多个全连接层。以及上述将上述历史浏览创意图像输入至预先训练的图编码模型，以生成历史图像编码向量，可以包括以下步骤：Optionally, the graph coding model includes: a residual network model and a plurality of fully connected layers. And the step of inputting the historical browsing creative image into a pre-trained graph coding model to generate a historical image coding vector may include the following steps:
第一步，将上述历史浏览创意图像输入至上述残差网络(ResNets，Residual Networks)模型，得到模型输出结果。The first step is to input the above historical browsing creative images into the above residual network (ResNets, Residual Networks) model to obtain the model output results.
第二步，将上述模型输出结果输入至上述多个全连接层，得到上 述历史图像编码向量。其中，多个全连接层可以是串行连接的多个全连接层。The second step is to input the output results of the above model into the above multiple fully connected layers to obtain the above The historical image coding vector is described. The multiple fully connected layers can be multiple fully connected layers connected in series.
步骤303，对上述用户稀疏特征信息集中的每个用户稀疏特征信息进行信息编码，以生成用户稀疏特征向量，得到用户稀疏特征向量集，以及对上述物品稀疏特征信息集中的每个物品稀疏特征信息进行信息编码，以生成物品稀疏特征向量，得到物品稀疏特征向量集。Step 303: Encode each user sparse feature information in the above user sparse feature information set to generate a user sparse feature vector, and obtain a user sparse feature vector set; and encode each item sparse feature information in the above item sparse feature information set to generate an item sparse feature vector, and obtain an item sparse feature vector set.
在一些实施例中，上述执行主体可以对上述用户稀疏特征信息集中的每个用户稀疏特征信息进行信息编码，以生成用户稀疏特征向量，得到用户稀疏特征向量集，以及对上述物品稀疏特征信息集中的每个物品稀疏特征信息进行信息编码，以生成物品稀疏特征向量，得到物品稀疏特征向量集。其中，用户稀疏特征向量可以表征用户稀疏特征信息的特征信息。物品稀疏特征向量可以表征物品稀疏特征信息的特征信息。In some embodiments, the execution subject may encode each user sparse feature information in the user sparse feature information set to generate a user sparse feature vector, and obtain a user sparse feature vector set, and may encode each item sparse feature information in the item sparse feature information set to generate an item sparse feature vector, and obtain an item sparse feature vector set. The user sparse feature vector may represent feature information of the user sparse feature information. The item sparse feature vector may represent feature information of the item sparse feature information.
作为示例，上述执行主体可以将用户稀疏特征信息集中的每个用户稀疏特征信息输入至Bert编码模型，以生成用户稀疏特征向量，得到用户稀疏特征向量集。同样的，上述执行主体可以将上述物品稀疏特征信息集中的每个物品稀疏特征信息输入至Bert编码模型，以生成物品稀疏特征向量，得到物品稀疏特征向量集。As an example, the execution subject may input each user sparse feature information in the user sparse feature information set into the Bert encoding model to generate a user sparse feature vector and obtain a user sparse feature vector set. Similarly, the execution subject may input each item sparse feature information in the item sparse feature information set into the Bert encoding model to generate an item sparse feature vector and obtain an item sparse feature vector set.
步骤304，对上述历史图像编码向量序列中的各个历史图像编码向量进行视觉偏好调整，得到调整后历史图像编码向量序列。Step 304 , performing visual preference adjustment on each historical image coding vector in the above historical image coding vector sequence to obtain an adjusted historical image coding vector sequence.
在一些实施例中，上述执行主体可以对上述历史图像编码向量序列中的各个历史图像编码向量进行视觉偏好调整，得到调整后历史图像编码向量序列。其中，调整后历史图像编码向量序列所包括的各个调整后历史图像编码向量对应的视觉偏好信息与历史图像编码向量序列包括的各个历史图像编码向量对应的视觉偏好信息相同。视觉偏好信息可以表征用户的视觉偏好特征。具体地，用户的视频偏好特征可以是但不限于以下之一：用户唯美视频偏好特征，用户搞笑视频偏好特征。各个调整后历史图像编码向量对应的视觉偏好信息可以是各个调整后历史图像编码向量所体现的用户的视觉偏好特征。各个历史图像编码向量对应的视觉偏好信息可以是各个历史图像编码向量所体现 的用户的视觉偏好特征。上述历史图像编码向量序列所包括的历史图像编码向量的数目与调整后历史图像编码向量序列所包括的调整后历史图像编码向量的数目相同。In some embodiments, the execution subject may perform visual preference adjustment on each historical image coding vector in the historical image coding vector sequence to obtain an adjusted historical image coding vector sequence. Among them, the visual preference information corresponding to each adjusted historical image coding vector included in the adjusted historical image coding vector sequence is the same as the visual preference information corresponding to each historical image coding vector included in the historical image coding vector sequence. The visual preference information may characterize the visual preference characteristics of the user. Specifically, the video preference characteristics of the user may be, but are not limited to, one of the following: a user's aesthetic video preference characteristics, a user's funny video preference characteristics. The visual preference information corresponding to each adjusted historical image coding vector may be the user's visual preference characteristics reflected by each adjusted historical image coding vector. The visual preference information corresponding to each historical image coding vector may be the visual preference characteristics reflected by each historical image coding vector. The number of historical image coding vectors included in the above historical image coding vector sequence is the same as the number of adjusted historical image coding vectors included in the adjusted historical image coding vector sequence.
作为示例，上述执行主体可以直接将历史图像编码向量序列中的各个历史图像编码向量输入至Transformer模型，以生成调整后历史图像编码向量序列。As an example, the above-mentioned execution entity can directly input each historical image coding vector in the historical image coding vector sequence into the Transformer model to generate an adjusted historical image coding vector sequence.
需要说明的是，调整后历史图像编码向量序列所包括的各个调整后历史图像编码向量所能体现的视觉偏好特征强于历史图像编码向量序列包括的各个历史图像编码向量所能体现的视觉偏好特征。It should be noted that the visual preference characteristics that can be reflected by each adjusted historical image coding vector included in the adjusted historical image coding vector sequence are stronger than the visual preference characteristics that can be reflected by each historical image coding vector included in the historical image coding vector sequence.
在一些实施例的一些可选的实现方式中，上述对上述历史图像编码向量序列中的各个历史图像编码向量进行视觉偏好调整，得到调整后历史图像编码向量序列，可以包括以下步骤：In some optional implementations of some embodiments, the above-mentioned performing visual preference adjustment on each historical image coding vector in the above-mentioned historical image coding vector sequence to obtain the adjusted historical image coding vector sequence may include the following steps:
第一步，确定上述历史图像编码向量序列对应的视觉偏好信息，作为目标视觉偏好信息。The first step is to determine the visual preference information corresponding to the above historical image coding vector sequence as the target visual preference information.
作为示例，上述执行主体可以将上述历史图像编码向量序列输入至Seq2Seq(Sequence to Sequence，序列到序列)模型，以输出目标视觉偏好信息。As an example, the above-mentioned execution entity can input the above-mentioned historical image encoding vector sequence into a Seq2Seq (Sequence to Sequence) model to output target visual preference information.
第二步，根据上述目标视觉偏好信息，对上述历史图像编码向量序列中的各个历史图像编码向量进行调整，得到调整后历史图像编码向量序列。In the second step, each historical image coding vector in the historical image coding vector sequence is adjusted according to the target visual preference information to obtain an adjusted historical image coding vector sequence.
作为示例，上述执行主体可以将目标视觉偏好信息和上述历史图像编码向量序列中的各个历史图像编码向量输入至生成式与对抗式神经网络(GAN，Generative Adversarial Network)模型，得到调整后历史图像编码向量序列。As an example, the above-mentioned execution entity can input the target visual preference information and each historical image coding vector in the above-mentioned historical image coding vector sequence into a generative adversarial neural network (GAN) model to obtain an adjusted historical image coding vector sequence.
步骤305，根据上述调整后历史图像编码向量序列、上述主图编码向量、上述用户稀疏特征向量集和上述物品稀疏特征向量集，利用多头注意力机制模型，生成待推送给上述目标用户的、上述目标推荐物品对应的推荐图像集。Step 305, based on the adjusted historical image coding vector sequence, the main image coding vector, the user sparse feature vector set and the item sparse feature vector set, a multi-head attention mechanism model is used to generate a recommended image set corresponding to the target recommended item to be pushed to the target user.
在一些实施例中，上述执行主体可以根据上述调整后历史图像编码向量序列、上述主图编码向量、上述用户稀疏特征向量集和上述物 品稀疏特征向量集，利用多头注意力机制(Multi-head-attention)模型，生成待推送给上述目标用户的、上述目标推荐物品对应的推荐图像集。其中，上述推荐图像集为上述创意图像特征信息集对应创意图像集的图像子集。。其中，多头注意力机制模型可以学习的多样化特征信息可以包括但不限于以下至少一项：调整后历史图像编码向量序列中的各个调整后历史图像编码向量之间的向量关联关系，用户稀疏特征向量集中的各个用户稀疏特征向量之间的向量关联关系，物品稀疏特征向量集中的各个物品稀疏特征向量之间的向量关联关系，调整后历史图像编码向量序列与主图编码向量之间的向量关联关系，调整后历史图像编码向量序列与调整后历史图像编码向量序列之间的向量关联关系，调整后历史图像编码向量序列与物品稀疏特征向量集之间的向量关联关系，主图编码向量与用户稀疏特征向量集之间的向量关联关系，主图编码向量与物品稀疏特征向量集之间的向量关联关系，用户稀疏特征向量集与物品稀疏特征向量集之间的向量关联关系。In some embodiments, the execution subject may perform the above-mentioned operation according to the above-mentioned adjusted historical image coding vector sequence, the above-mentioned main image coding vector, the above-mentioned user sparse feature vector set and the above-mentioned object The multi-head attention mechanism model generates a recommended image set corresponding to the target recommended item to be pushed to the target user. The recommended image set is an image subset of the creative image set corresponding to the creative image feature information set. The diversified feature information that can be learned by the multi-head attention mechanism model may include but is not limited to at least one of the following: the vector correlation relationship between each adjusted historical image coding vector in the adjusted historical image coding vector sequence, the vector correlation relationship between each user sparse feature vector in the user sparse feature vector set, the vector correlation relationship between each item sparse feature vector in the item sparse feature vector set, the vector correlation relationship between the adjusted historical image coding vector sequence and the main image coding vector, the vector correlation relationship between the adjusted historical image coding vector sequence and the adjusted historical image coding vector sequence, the vector correlation relationship between the adjusted historical image coding vector sequence and the item sparse feature vector set, the vector correlation relationship between the main image coding vector and the user sparse feature vector set, the vector correlation relationship between the main image coding vector and the item sparse feature vector set, and the vector correlation relationship between the user sparse feature vector set and the item sparse feature vector set.
作为示例，上述执行主体可以将上述调整后历史图像编码向量序列、上述主图编码向量、上述用户稀疏特征向量集和上述物品稀疏特征向量集输入至多头注意力机制模型，生成待推送给上述目标用户的、上述目标推荐物品对应的推荐图像集。As an example, the above-mentioned execution entity can input the above-mentioned adjusted historical image encoding vector sequence, the above-mentioned main image encoding vector, the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set into the multi-head attention mechanism model to generate a recommended image set corresponding to the above-mentioned target recommended item to be pushed to the above-mentioned target user.
本公开的上述各个实施例具有如下有益效果：通过本公开的一些实施例的推荐信息生成方法，生成了精准的推荐信息，得到了较好的推荐效果。具体来说，造成相关的推荐效果不佳的原因在于：仅利用针对目标用户的历史浏览创意图像序列，对用户素材偏好模型进行训练，使得用户素材偏好模型所能学习到的特征信息有限，以致用户素材偏好模型不够精准。侧面导致推荐效果不佳。基于此，本公开的一些实施例的推荐信息生成方法，首先，获取目标用户的历史浏览创意图像序列、上述目标用户的用户稀疏特征信息集、目标推荐物品的主图像和上述目标推荐物品的物品稀疏特征信息集，其中，上述物品稀疏特征信息集包括针对上述目标推荐物品的创意图像特征信息集。以用于后续获取更多的特征信息，便于后续生成更为精准的推荐信息(即，多个推荐创意图像分数和推荐主图像分数)。然后，对上述主图 像进行图编码处理，得到主图编码向量，以及对上述历史浏览创意图像序列中的每个历史浏览创意图像进行图编码处理，以生成历史图像编码向量，得到历史图像编码向量序列。在这里，对主图像进行图编码处理，以提取主图像的特征信息。除此之外，相对于通过主图像来生成推荐信息，利用主图编码向量来生成推荐信息可以有效地解决由于图像像素维度较大而带来的计算量较大的问题。同样地，相对于通过历史浏览创意图像序列来进行视觉偏好调整，对历史图像编码向量序列进行视觉偏好调整可以有效解决计算量较大的问题。接着，对上述用户稀疏特征信息集中的每个用户稀疏特征信息进行信息编码，以转换成向量形式，便于用户稀疏特征的特征信息的使用。同样地，对上述物品稀疏特征信息集中的每个物品稀疏特征信息进行信息编码，以转换成向量形式，便于物品稀疏特征的特征信息的使用。进而，对上述历史图像编码向量序列中的各个历史图像编码向量进行视觉偏好调整，以使得调整后历史图像编码向量序列所能体现的视觉偏好更为明显，有助于后续生成更为精准的推荐信息。最后，通过多头注意力机制模型，可以提取针对上述调整后历史图像编码向量序列、上述主图编码向量、上述用户稀疏特征向量集和上述物品稀疏特征向量集的多方面的特征信息。由此，所生成的推荐信息(即，推荐图像集)更为精准。The above-mentioned various embodiments of the present disclosure have the following beneficial effects: through the recommendation information generation method of some embodiments of the present disclosure, accurate recommendation information is generated, and a good recommendation effect is obtained. Specifically, the reason for the poor recommendation effect is that only the historical browsing creative image sequence of the target user is used to train the user material preference model, so that the feature information that the user material preference model can learn is limited, so that the user material preference model is not accurate enough. This indirectly leads to poor recommendation effect. Based on this, the recommendation information generation method of some embodiments of the present disclosure first obtains the historical browsing creative image sequence of the target user, the user sparse feature information set of the above target user, the main image of the target recommended item, and the item sparse feature information set of the above target recommended item, wherein the above item sparse feature information set includes the creative image feature information set for the above target recommended item. This is used to subsequently obtain more feature information, so as to facilitate the subsequent generation of more accurate recommendation information (i.e., multiple recommended creative image scores and recommended main image scores). Then, the above main image The image is subjected to image coding processing to obtain the main image coding vector, and each historical browsing creative image in the above-mentioned historical browsing creative image sequence is subjected to image coding processing to generate a historical image coding vector to obtain a historical image coding vector sequence. Here, the main image is subjected to image coding processing to extract the feature information of the main image. In addition, compared with generating recommendation information through the main image, generating recommendation information using the main image coding vector can effectively solve the problem of large amount of calculation due to large image pixel dimensions. Similarly, compared with adjusting visual preference through the historical browsing creative image sequence, adjusting visual preference of the historical image coding vector sequence can effectively solve the problem of large amount of calculation. Next, information encoding is performed on each user sparse feature information in the above-mentioned user sparse feature information set to convert it into a vector form, so as to facilitate the use of the feature information of the user sparse feature. Similarly, information encoding is performed on each item sparse feature information in the above-mentioned item sparse feature information set to convert it into a vector form, so as to facilitate the use of the feature information of the item sparse feature. Furthermore, visual preference adjustment is performed on each historical image coding vector in the above-mentioned historical image coding vector sequence, so that the visual preference reflected by the adjusted historical image coding vector sequence is more obvious, which is helpful for the subsequent generation of more accurate recommendation information. Finally, through the multi-head attention mechanism model, various feature information of the adjusted historical image encoding vector sequence, the main image encoding vector, the user sparse feature vector set and the item sparse feature vector set can be extracted. As a result, the generated recommendation information (i.e., the recommended image set) is more accurate.
参考图4，示出了根据本公开的推荐信息生成方法的另一些实施例的流程400。该推荐信息生成方法，包括以下步骤：Referring to FIG4 , a process 400 of another embodiment of a method for generating recommendation information according to the present disclosure is shown. The method for generating recommendation information includes the following steps:
步骤401，获取目标用户的历史浏览创意图像序列、上述目标用户的用户稀疏特征信息集、目标推荐物品的主图像和上述目标推荐物品的物品稀疏特征信息集。Step 401, obtaining a target user's historical browsing creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item.
步骤402，对上述主图像进行图编码处理，得到主图编码向量，以及对上述历史浏览创意图像序列中的每个历史浏览创意图像进行图编码处理，以生成历史图像编码向量，得到历史图像编码向量序列。Step 402, performing image coding processing on the main image to obtain a main image coding vector, and performing image coding processing on each historically browsed creative image in the historically browsed creative image sequence to generate a historical image coding vector to obtain a historical image coding vector sequence.
步骤403，对上述用户稀疏特征信息集中的每个用户稀疏特征信息进行信息编码，以生成用户稀疏特征向量，得到用户稀疏特征向量 集，以及对上述物品稀疏特征信息集中的每个物品稀疏特征信息进行信息编码，以生成物品稀疏特征向量，得到物品稀疏特征向量集。Step 403: Encode the sparse feature information of each user in the above user sparse feature information set to generate a user sparse feature vector, and obtain the user sparse feature vector A set is obtained, and information encoding is performed on each item sparse feature information in the item sparse feature information set to generate an item sparse feature vector, thereby obtaining an item sparse feature vector set.
步骤404，对上述历史图像编码向量序列中的各个历史图像编码向量进行视觉偏好调整，得到调整后历史图像编码向量序列。Step 404 , performing visual preference adjustment on each historical image coding vector in the above historical image coding vector sequence to obtain an adjusted historical image coding vector sequence.
在一些实施例中，步骤401-404的具体实现及其所带来的技术效果，可以参考图3对应的实施例中的步骤301-304，在此不再赘述。In some embodiments, the specific implementation of steps 401-404 and the technical effects brought about by them can refer to steps 301-304 in the embodiment corresponding to FIG. 3, and will not be repeated here.
步骤405，将上述用户稀疏特征向量集和上述物品稀疏特征向量集进行向量拼接，得到拼接稀疏特征向量。Step 405 , concatenate the user sparse feature vector set and the item sparse feature vector set to obtain a concatenated sparse feature vector.
在一些实施例中，执行主体(例如图1所示的电子设备101)可以将上述用户稀疏特征向量集和上述物品稀疏特征向量集进行向量拼接，得到拼接稀疏特征向量。In some embodiments, the execution entity (eg, the electronic device 101 shown in FIG. 1 ) may perform vector concatenation on the user sparse feature vector set and the item sparse feature vector set to obtain a concatenated sparse feature vector.
步骤406，将上述调整后历史图像编码向量序列、上述主图编码向量、上述拼接稀疏特征向量输入至上述多头注意力机制模型，得到特征信息融合向量。Step 406, input the adjusted historical image coding vector sequence, the main image coding vector, and the spliced sparse feature vector into the multi-head attention mechanism model to obtain a feature information fusion vector.
在一些实施例中，上述执行主体可以将上述调整后历史图像编码向量序列、上述主图编码向量、上述拼接稀疏特征向量输入至上述多头注意力机制模型，得到特征信息融合向量。特征信息融合向量是由多种特征信息融合得到的。其中，特征信息融合向量可以包括但不限于以下至少一项：调整后历史图像编码向量序列中各个历史图像编码向量间的向量关系信息，调整后历史图像编码向量序列与上述主图编码向量之间的向量关系信息，拼接稀疏特征向量与调整后历史图像编码向量序列之间的向量关系信息，拼接稀疏特征向量与主图编码向量之间的向量关系信息。In some embodiments, the execution subject may input the adjusted historical image coding vector sequence, the main image coding vector, and the spliced sparse feature vector into the multi-head attention mechanism model to obtain a feature information fusion vector. The feature information fusion vector is obtained by fusing multiple feature information. The feature information fusion vector may include but is not limited to at least one of the following: vector relationship information between each historical image coding vector in the adjusted historical image coding vector sequence, vector relationship information between the adjusted historical image coding vector sequence and the main image coding vector, vector relationship information between the spliced sparse feature vector and the adjusted historical image coding vector sequence, and vector relationship information between the spliced sparse feature vector and the main image coding vector.
除此之外，多头注意力机制模型可以学习的多样化特征信息还可以包括：稀疏特征信息(即用户稀疏特征向量集和物品稀疏特征向量集)的重要程度信息和素材内容信息(即上述调整后历史图像编码向量序列和上述主图编码向量)的重要程度信息。In addition, the diversified feature information that can be learned by the multi-head attention mechanism model can also include: the importance information of sparse feature information (i.e., user sparse feature vector sets and item sparse feature vector sets) and the importance information of material content information (i.e., the above-mentioned adjusted historical image encoding vector sequence and the above-mentioned main image encoding vector).
步骤407，将上述拼接稀疏特征向量输入至全连接模型，得到输出向量。Step 407: input the concatenated sparse feature vector into a fully connected model to obtain an output vector.
在一些实施例中，上述执行主体可以将上述拼接稀疏特征向量输 入至全连接模型，得到输出向量。其中，上述全连接模型可以包括多个全连接层。In some embodiments, the execution entity may input the concatenated sparse feature vector into The fully connected model can include multiple fully connected layers.
步骤408，利用预设损失函数，生成上述推荐图像集。在一些实施例中，上述执行主体可以利用预设损失函数，通过各种方式来生成上述推荐图像集。例如，预设损失函数可以是平方损失函数(quadratic loss function)。Step 408, using a preset loss function to generate the recommended image set. In some embodiments, the execution subject may use a preset loss function to generate the recommended image set in various ways. For example, the preset loss function may be a quadratic loss function.
在一些实施例的一些可选的实现方式中，上述利用预设损失函数，生成上述推荐图像集，可以包括以下步骤：In some optional implementations of some embodiments, the above-mentioned generation of the above-mentioned recommended image set by using a preset loss function may include the following steps:
第一步，利用上述预设损失函数，生成针对上述拼接向量的创意图像分数集。其中，创意图像分数表征上述目标用户对上述目标推荐物品的创意图像的感兴趣程度。上述创意图像分数集中的创意图像分数集与创意图像集中的创意图像存在一一对应的。In the first step, the preset loss function is used to generate a creative image score set for the splicing vector. The creative image score represents the interest of the target user in the creative image of the target recommended item. The creative image score set in the creative image score set corresponds to the creative image in the creative image set.
作为示例，上述执行主体可以将上述拼接向量输入至预设损失函数，得到针对上述拼接向量的创意图像分数集。As an example, the execution entity may input the stitching vector into a preset loss function to obtain a creative image score set for the stitching vector.
第二步，根据上述调整后历史图像编码向量序列、上述主图编码向量、上述用户稀疏特征向量集和上述物品稀疏特征向量集，利用多头注意力机制模型，生成上述目标用户对应的推荐物品分数。其中，推荐物品分数可以表征目标用户对上述目标推荐物品的喜好程度。The second step is to generate the recommended item score corresponding to the target user using the multi-head attention mechanism model based on the adjusted historical image coding vector sequence, the main image coding vector, the user sparse feature vector set and the item sparse feature vector set. The recommended item score can represent the target user's preference for the target recommended item.
作为示例，上述执行主体可以将上述调整后历史图像编码向量序列、上述主图编码向量、上述用户稀疏特征向量集和上述物品稀疏特征向量集输入至多头注意力机制模型，生成上述目标用户对应的推荐物品分数。As an example, the above-mentioned execution entity can input the above-mentioned adjusted historical image encoding vector sequence, the above-mentioned main image encoding vector, the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set into the multi-head attention mechanism model to generate the recommended item score corresponding to the above-mentioned target user.
第三步，响应于确定上述推荐物品分数大于预定推荐物品数值，上述执行主体可以确定创意图像分数集中分数值大于预定推荐创意图像数值的创意图像分数，得到创意图像分数子集。In the third step, in response to determining that the recommended item score is greater than the predetermined recommended item value, the execution entity may determine the creative image scores in the creative image score set whose score values are greater than the predetermined recommended creative image value, and obtain a creative image score subset.
例如，预定推荐物品数值为75分。For example, the value of the predetermined recommended item is 75 points.
第四步，上述执行主体可以将上述创意图像分数子集对应的创意图像集确定为上述推荐图像集。In a fourth step, the execution entity may determine the creative image set corresponding to the creative image score subset as the recommended image set.
可选地，步骤还包括：Optionally, the steps further include:
第一步，根据上述调整后历史图像编码向量序列、上述主图编码 向量、上述用户稀疏特征向量集和上述物品稀疏特征向量集，利用多头注意力机制模型，生成上述目标推荐物品对应的推荐主图像分数。其中，上述推荐主图像分数表征上述目标用户对上述主图像的感兴趣程度。The first step is to use the above adjusted historical image coding vector sequence and the above main image coding vector sequence. The vector, the user sparse feature vector set and the item sparse feature vector set are used to generate a recommended main image score corresponding to the target recommended item using a multi-head attention mechanism model. The recommended main image score represents the interest of the target user in the main image.
作为示例，上述执行主体可以将上述调整后历史图像编码向量序列、上述主图编码向量、上述用户稀疏特征向量集和上述物品稀疏特征向量集输入至多头注意力机制模型，生成上述目标推荐物品对应的推荐主图像分数。As an example, the above-mentioned execution entity can input the above-mentioned adjusted historical image encoding vector sequence, the above-mentioned main image encoding vector, the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set into the multi-head attention mechanism model to generate the recommended main image score corresponding to the above-mentioned target recommended item.
第二步，响应于确定上述推荐物品分数大于预定推荐物品数值、且上述推荐主图像分数大于预定推荐主图像数值，将上述主图像推送至上述目标用户对应的终端。其中，目标用户对应的终端可以是显示终端。In the second step, in response to determining that the recommended item score is greater than a predetermined recommended item value and the recommended main image score is greater than a predetermined recommended main image value, the main image is pushed to a terminal corresponding to the target user. The terminal corresponding to the target user may be a display terminal.
例如，预定推荐物品数值可以是70。预定推荐主图像数值可以是75。For example, the predetermined recommended item value may be 70. The predetermined recommended main image value may be 75.
从图4中可以看出，与图3对应的一些实施例的描述相比，图4对应的一些实施例中的推荐信息生成方法的流程400，利用多头注意力机制模型，可以学习针对稀疏特征信息(即用户稀疏特征向量集和物品稀疏特征向量集)的重要程度信息和素材内容信息(即上述调整后历史图像编码向量序列和上述主图编码向量)的重要程度信息的多样化特征信息，以此可以生成更为精准的创意图像集。It can be seen from Figure 4 that compared with the description of some embodiments corresponding to Figure 3, the process 400 of the recommendation information generation method in some embodiments corresponding to Figure 4, using a multi-head attention mechanism model, can learn the diversified feature information of the importance information of sparse feature information (i.e., user sparse feature vector set and item sparse feature vector set) and the importance information of material content information (i.e., the above-mentioned adjusted historical image coding vector sequence and the above-mentioned main image coding vector), so as to generate a more accurate creative image set.
参考图5，作为对上述各图所示方法的实现，本公开提供了一种推荐信息生成装置的一些实施例，这些装置实施例与图3所示的那些方法实施例相对应，该装置具体可以应用于各种电子设备中。Referring to FIG. 5 , as an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a device for generating recommendation information. These device embodiments correspond to the method embodiments shown in FIG. 3 , and the device can be specifically applied to various electronic devices.
如图5所示，一种推荐信息生成装置500包括：获取单元501、图编码单元502、信息编码单元503、调整单元504和生成单元505。其中，获取单元501，被配置成获取目标用户的历史浏览创意图像序列、上述目标用户的用户稀疏特征信息集、目标推荐物品的主图像和上述目标推荐物品的物品稀疏特征信息集，其中，上述物品稀疏特征信息集包括针对上述目标推荐物品的创意图像特征信息集；图编码单 元502，被配置成对上述主图像进行图编码处理，得到主图编码向量，以及对上述历史浏览创意图像序列中的每个历史浏览创意图像进行图编码处理，以生成历史图像编码向量，得到历史图像编码向量序列；信息编码单元503，被配置成对上述用户稀疏特征信息集中的每个用户稀疏特征信息进行信息编码，以生成用户稀疏特征向量，得到用户稀疏特征向量集，以及对上述物品稀疏特征信息集中的每个物品稀疏特征信息进行信息编码，以生成物品稀疏特征向量，得到物品稀疏特征向量集；调整单元504，被配置成对上述历史图像编码向量序列中的各个历史图像编码向量进行视觉偏好调整，得到调整后历史图像编码向量序列；生成单元505，被配置成根据上述调整后历史图像编码向量序列、上述主图编码向量、上述用户稀疏特征向量集和上述物品稀疏特征向量集，利用多头注意力机制模型，生成待推送给上述目标用户的、上述目标推荐物品对应的推荐图像集。As shown in FIG5 , a recommendation information generating device 500 includes: an acquisition unit 501, a graph encoding unit 502, an information encoding unit 503, an adjustment unit 504, and a generation unit 505. The acquisition unit 501 is configured to acquire a target user's historical browsing creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item, wherein the item sparse feature information set includes a creative image feature information set for the target recommended item; the graph encoding unit 502 is configured to acquire a target user's historical browsing creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item; Element 502 is configured to perform image coding processing on the above-mentioned main image to obtain a main image coding vector, and to perform image coding processing on each historical browsing creative image in the above-mentioned historical browsing creative image sequence to generate a historical image coding vector to obtain a historical image coding vector sequence; information coding unit 503 is configured to perform information coding on each user sparse feature information in the above-mentioned user sparse feature information set to generate a user sparse feature vector to obtain a user sparse feature vector set, and to perform information coding on each item sparse feature information in the above-mentioned item sparse feature information set to generate an item sparse feature vector to obtain an item sparse feature vector set; adjustment unit 504 is configured to perform visual preference adjustment on each historical image coding vector in the above-mentioned historical image coding vector sequence to obtain an adjusted historical image coding vector sequence; generation unit 505 is configured to generate a recommended image set corresponding to the above-mentioned target recommended item to be pushed to the above-mentioned target user based on the above-mentioned adjusted historical image coding vector sequence, the above-mentioned main image coding vector, the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set, using a multi-head attention mechanism model.
在一些实施例的一些可选的实现方式中，上述装置500中的调整单元504可以被配置成：确定上述历史图像编码向量序列对应的视觉偏好信息，作为目标视觉偏好信息；根据上述目标视觉偏好信息，对上述历史图像编码向量序列中的各个历史图像编码向量进行调整，得到调整后历史图像编码向量序列。In some optional implementations of some embodiments, the adjustment unit 504 in the above-mentioned device 500 can be configured to: determine the visual preference information corresponding to the above-mentioned historical image coding vector sequence as the target visual preference information; and adjust each historical image coding vector in the above-mentioned historical image coding vector sequence according to the above-mentioned target visual preference information to obtain an adjusted historical image coding vector sequence.
在一些实施例的一些可选的实现方式中，上述装置500中的生成单元505可以被配置成：将上述用户稀疏特征向量集和上述物品稀疏特征向量集进行向量拼接，得到拼接稀疏特征向量；将上述调整后历史图像编码向量序列、上述主图编码向量、上述拼接稀疏特征向量输入至上述多头注意力机制模型，得到特征信息融合向量；将上述拼接稀疏特征向量输入至全连接模型，得到输出向量；将上述特征信息融合向量和上述输出向量进行拼接，得到拼接向量；利用预设损失函数，生成上述推荐图像集。In some optional implementations of some embodiments, the generation unit 505 in the above-mentioned device 500 can be configured to: perform vector splicing on the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set to obtain a spliced sparse feature vector; input the above-mentioned adjusted historical image encoding vector sequence, the above-mentioned main image encoding vector, and the above-mentioned spliced sparse feature vector into the above-mentioned multi-head attention mechanism model to obtain a feature information fusion vector; input the above-mentioned spliced sparse feature vector into the fully connected model to obtain an output vector; splice the above-mentioned feature information fusion vector and the above-mentioned output vector to obtain a spliced vector; and use a preset loss function to generate the above-mentioned recommended image set.
在一些实施例的一些可选的实现方式中，上述图编码模型包括：残差网络模型和多个全连接层；上述装置500中的图编码单元502可以被配置成：将上述历史浏览创意图像输入至上述残差网络模型，得到模型输出结果；将上述模型输出结果输入至上述多个全连接层，得 到上述历史图像编码向量。In some optional implementations of some embodiments, the graph coding model includes: a residual network model and a plurality of fully connected layers; the graph coding unit 502 in the apparatus 500 may be configured to: input the historical browsing creative image into the residual network model to obtain a model output result; input the model output result into the plurality of fully connected layers to obtain to the above historical image encoding vector.
在一些实施例的一些可选的实现方式中，上述装置500中的生成单元505可以被配置成：利用上述预设损失函数，生成针对上述拼接向量的创意图像分数集；根据上述调整后历史图像编码向量序列、上述主图编码向量、上述用户稀疏特征向量集和上述物品稀疏特征向量集，利用多头注意力机制模型，生成上述目标推荐物品对应的推荐物品分数；响应于确定上述推荐物品分数大于预定推荐物品数值，确定上述创意图像分数集中分数值大于预定推荐创意图像数值的创意图像分数，得到创意图像分数子集；将上述创意图像分数子集对应的创意图像集确定为上述推荐图像集。In some optional implementations of some embodiments, the generation unit 505 in the above-mentioned device 500 can be configured to: generate a creative image score set for the above-mentioned splicing vector using the above-mentioned preset loss function; generate a recommended item score corresponding to the above-mentioned target recommended item using a multi-head attention mechanism model based on the above-mentioned adjusted historical image coding vector sequence, the above-mentioned main image coding vector, the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set; in response to determining that the above-mentioned recommended item score is greater than a predetermined recommended item value, determine the creative image score in the above-mentioned creative image score set whose score value is greater than a predetermined recommended creative image value, and obtain a creative image score subset; determine the creative image set corresponding to the above-mentioned creative image score subset as the above-mentioned recommended image set.
在一些实施例的一些可选的实现方式中，上述装置500中的生成单元505可以被配置成：根据上述调整后历史图像编码向量序列、上述主图编码向量、上述用户稀疏特征向量集和上述物品稀疏特征向量集，利用多头注意力机制模型，生成上述目标推荐物品对应的推荐主图像分数；响应于确定上述推荐物品分数大于预定推荐物品数值、且上述推荐主图像分数大于预定推荐主图像数值，将上述主图像推送至上述目标用户对应的终端。In some optional implementations of some embodiments, the generation unit 505 in the above-mentioned device 500 can be configured to: generate a recommended main image score corresponding to the above-mentioned target recommended item based on the above-mentioned adjusted historical image encoding vector sequence, the above-mentioned main image encoding vector, the above-mentioned user sparse feature vector set and the above-mentioned item sparse feature vector set using a multi-head attention mechanism model; in response to determining that the above-mentioned recommended item score is greater than a predetermined recommended item value and the above-mentioned recommended main image score is greater than a predetermined recommended main image value, push the above-mentioned main image to the terminal corresponding to the above-mentioned target user.
可以理解的是，该装置500中记载的诸单元与参考图3描述的方法中的各个步骤相对应。由此，上文针对方法描述的操作、特征以及产生的有益效果同样适用于装置500及其中包含的单元，在此不再赘述。It is understandable that the units recorded in the device 500 correspond to the steps in the method described with reference to Figure 3. Therefore, the operations, features and beneficial effects described above for the method are also applicable to the device 500 and the units contained therein, and will not be repeated here.
下面参考图6，其示出了适于用来实现本公开的一些实施例的电子设备(例如图1中的电子设备101)600的结构示意图。图6示出的电子设备仅仅是一个示例，不应对本公开的实施例的功能和使用范围带来任何限制。Referring to FIG6, a schematic diagram of an electronic device (such as the electronic device 101 in FIG1) 600 suitable for implementing some embodiments of the present disclosure is shown. The electronic device shown in FIG6 is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure.
如图6所示，电子设备600可以包括处理装置(例如中央处理器、图形处理器等)601，其可以根据存储在只读存储器(ROM)602中的程序或者从存储装置608加载到随机访问存储器(RAM)603中的程序而执行各种适当的动作和处理。在RAM 603中，还存储有电子设备 600操作所需的各种程序和数据。处理装置601、ROM 602以及RAM603通过总线604彼此相连。输入/输出(I/O)接口605也连接至总线604。As shown in FIG6 , the electronic device 600 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which may perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 to a random access memory (RAM) 603. The RAM 603 also stores the electronic device Various programs and data required for the operation of the bus 600. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input/output (I/O) interface 605 is also connected to the bus 604.
通常，以下装置可以连接至I/O接口605：包括例如触摸屏、触摸板、键盘、鼠标、摄像头、麦克风、加速度计、陀螺仪等的输入装置606；包括例如液晶显示器(LCD)、扬声器、振动器等的输出装置607；包括例如磁带、硬盘等的存储装置608；以及通信装置609。通信装置609可以允许电子设备600与其他设备进行无线或有线通信以交换数据。虽然图6示出了具有各种装置的电子设备600，但是应理解的是，并不要求实施或具备所有示出的装置。可以替代地实施或具备更多或更少的装置。图6中示出的每个方框可以代表一个装置，也可以根据需要代表多个装置。Typically, the following devices may be connected to the I/O interface 605: input devices 606 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 608 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 609. The communication device 609 may allow the electronic device 600 to communicate wirelessly or wired with other devices to exchange data. Although FIG. 6 shows an electronic device 600 with various devices, it should be understood that it is not required to implement or have all the devices shown. More or fewer devices may be implemented or have alternatively. Each box shown in FIG. 6 may represent one device, or may represent multiple devices as needed.
特别地，根据本公开的一些实施例，上文参考流程图描述的过程可以被实现为计算机软件程序。例如，本公开的一些实施例包括一种计算机程序产品，其包括承载在计算机可读介质上的计算机程序，该计算机程序包含用于执行流程图所示的方法的程序代码。在这样的一些实施例中，该计算机程序可以通过通信装置609从网络上被下载和安装，或者从存储装置608被安装，或者从ROM 602被安装。在该计算机程序被处理装置601执行时，执行本公开的一些实施例的方法中限定的上述功能。In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 609, or installed from the storage device 608, or installed from the ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.
需要说明的是，本公开的一些实施例上述的计算机可读介质可以是计算机可读信号介质或者计算机可读存储介质或者是上述两者的任意组合。计算机可读存储介质例如可以是——但不限于——电、磁、光、电磁、红外线、或半导体的***、装置或器件，或者任意以上的组合。计算机可读存储介质的更具体的例子可以包括但不限于：具有至少一个导线的电连接、便携式计算机磁盘、硬盘、随机访问存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑磁盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。在本公开的一些实施例中，计算机可读存储介质可以是任何包含或存储程序的有形介质， 该程序可以被指令执行***、装置或者器件使用或者与其结合使用。而在本公开的一些实施例中，计算机可读信号介质可以包括在基带中或者作为载波一部分传播的数据信号，其中承载了计算机可读的程序代码。这种传播的数据信号可以采用多种形式，包括但不限于电磁信号、光信号或上述的任意合适的组合。计算机可读信号介质还可以是计算机可读存储介质以外的任何计算机可读介质，该计算机可读信号介质可以发送、传播或者传输用于由指令执行***、装置或者器件使用或者与其结合使用的程序。计算机可读介质上包含的程序代码可以用任何适当的介质传输，包括但不限于：电线、光缆、RF(射频)等等，或者上述的任意合适的组合。It should be noted that in some embodiments of the present disclosure, the above-mentioned computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with at least one wire, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program, The program can be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal can take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate or transmit a program for use by or in combination with an instruction execution system, device or device. The program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.
在一些实施方式中，客户端、服务器可以利用诸如HTTP(Hyper Text Transfer Protocol，超文本传输协议)之类的任何当前已知或未来研发的网络协议进行通信，并且可以与任意形式或介质的数字数据通信(例如，通信网络)互连。通信网络的示例包括局域网(“LAN”)，广域网(“WAN”)，网际网(例如，互联网)以及端对端网络(例如，ad hoc端对端网络)，以及任何当前已知或未来研发的网络。In some embodiments, the client and server may communicate using any currently known or future developed network protocol such as HTTP (Hyper Text Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.
上述计算机可读介质可以是上述电子设备中所包含的；也可以是单独存在，而未装配入该电子设备中。上述计算机可读介质承载有一个或者多个程序，当上述一个或者多个程序被该电子设备执行时，使得该电子设备：获取目标用户的历史浏览创意图像序列、上述目标用户的用户稀疏特征信息集、目标推荐物品的主图像和上述目标推荐物品的物品稀疏特征信息集，其中，上述物品稀疏特征信息集包括针对上述目标推荐物品的创意图像特征信息集；对上述主图像进行图编码处理，得到主图编码向量，以及对上述历史浏览创意图像序列中的每个历史浏览创意图像进行图编码处理，以生成历史图像编码向量，得到历史图像编码向量序列；对上述用户稀疏特征信息集中的每个用户稀疏特征信息进行信息编码，以生成用户稀疏特征向量，得到用户稀疏特征向量集，以及对上述物品稀疏特征信息集中的每个物品稀疏特征信息进行信息编码，以生成物品稀疏特征向量，得到物品稀疏特征向量集；对上述历史图像编码向量序列中的各个历史图像编码向量进 行视觉偏好调整，得到调整后历史图像编码向量序列；根据上述调整后历史图像编码向量序列、上述主图编码向量、上述用户稀疏特征向量集和上述物品稀疏特征向量集，利用多头注意力机制模型，生成待推送给上述目标用户的、上述目标推荐物品对应的推荐图像集。The computer-readable medium may be included in the electronic device; or it may exist independently without being installed in the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: obtains the target user's historical browsing creative image sequence, the target user's user sparse feature information set, the target recommended item's main image, and the target recommended item's item sparse feature information set, wherein the item sparse feature information set includes a creative image feature information set for the target recommended item; performs image coding processing on the main image to obtain a main image coding vector, and performs image coding processing on each historical browsing creative image in the historical browsing creative image sequence to generate a historical image coding vector to obtain a historical image coding vector sequence; performs information coding on each user sparse feature information in the user sparse feature information set to generate a user sparse feature vector to obtain a user sparse feature vector set, and performs information coding on each item sparse feature information in the item sparse feature information set to generate an item sparse feature vector to obtain an item sparse feature vector set; performs information coding on each historical image coding vector in the historical image coding vector sequence Perform visual preference adjustment to obtain an adjusted historical image coding vector sequence; based on the adjusted historical image coding vector sequence, the main image coding vector, the user sparse feature vector set and the item sparse feature vector set, use a multi-head attention mechanism model to generate a recommended image set corresponding to the target recommended item to be pushed to the target user.
可以以一种或多种程序设计语言或其组合来编写用于执行本公开的一些实施例的操作的计算机程序代码，上述程序设计语言包括面向对象的程序设计语言—诸如Java、Smalltalk、C++，还包括常规的过程式程序设计语言—诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络——包括局域网(LAN)或广域网(WAN)——连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
附图中的流程图和框图，图示了按照本公开各种实施例的***、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段、或代码的一部分，该模块、程序段、或代码的一部分包含至少一个用于实现规定的逻辑功能的可执行指令。也应当注意，在有些作为替换的实现中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个接连地表示的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图和/或流程图中的每个方框、以及框图和/或流程图中的方框的组合，可以用执行规定的功能或操作的专用的基于硬件的***来实现，或者可以用专用硬件与计算机指令的组合来实现。The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains at least one executable instruction for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and/or flow chart, and the combination of the square boxes in the block diagram and/or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
描述于本公开的一些实施例中的单元可以通过软件的方式实现，也可以通过硬件的方式来实现。所描述的单元也可以设置在处理器中，例如，可以描述为：一种处理器包括获取单元、图编码单元、信息编码单元、调整单元和生成单元。其中，这些单元的名称在某种情况下并不构成对该单元本身的限定，例如，获取单元还可以被描述为“获 取目标用户的历史浏览创意图像序列、上述目标用户的用户稀疏特征信息集、目标推荐物品的主图像和上述目标推荐物品的物品稀疏特征信息集的单元”。The units described in some embodiments of the present disclosure may be implemented by software or hardware. The units described may also be set in a processor. For example, they may be described as follows: a processor includes an acquisition unit, a graph encoding unit, an information encoding unit, an adjustment unit, and a generation unit. The names of these units do not, in some cases, limit the units themselves. For example, the acquisition unit may also be described as “acquisition unit”. Take a unit of a target user's historical browsing creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item.
本文中以上描述的功能可以至少部分地由至少一个硬件逻辑部件来执行。例如，非限制性地，可以使用的示范类型的硬件逻辑部件包括：现场可编程门阵列(FPGA)、专用集成电路(ASIC)、专用标准产品(ASSP)、片上***(SOC)、复杂可编程逻辑设备(CPLD)等等。The functions described above herein may be performed at least in part by at least one hardware logic component. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.
本公开的一些实施例还提供一种计算机程序产品，包括计算机程序，计算机程序在被处理器执行时实现上述的任一种推荐信息生成方法。Some embodiments of the present disclosure further provide a computer program product, including a computer program, which implements any of the above-mentioned recommendation information generation methods when executed by a processor.
以上描述仅为本公开的一些较佳实施例以及对所运用技术原理的说明。本领域技术人员应当理解，本公开的实施例中所涉及的发明范围，并不限于上述技术特征的特定组合而成的技术方案，同时也应涵盖在不脱离上述发明构思的情况下，由上述技术特征或其等同特征进行任意组合而形成的其它技术方案。例如上述特征与本公开的实施例中公开的(但不限于)具有类似功能的技术特征进行互相替换而形成的技术方案。 The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above invention concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure.

Claims (11)

  1. 一种推荐信息生成方法，包括：A method for generating recommendation information, comprising:
    获取目标用户的历史浏览创意图像序列、所述目标用户的用户稀疏特征信息集、目标推荐物品的主图像和所述目标推荐物品的物品稀疏特征信息集，其中，所述物品稀疏特征信息集包括针对所述目标推荐物品的创意图像特征信息集；Acquire a target user's historical browsing creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item, wherein the item sparse feature information set includes a creative image feature information set for the target recommended item;
    对所述主图像进行图编码处理，得到主图编码向量，以及对所述历史浏览创意图像序列中的每个历史浏览创意图像进行图编码处理，以生成历史图像编码向量，得到历史图像编码向量序列；Performing image coding processing on the main image to obtain a main image coding vector, and performing image coding processing on each historical browsing creative image in the historical browsing creative image sequence to generate a historical image coding vector to obtain a historical image coding vector sequence;
    对所述用户稀疏特征信息集中的每个用户稀疏特征信息进行信息编码，以生成用户稀疏特征向量，得到用户稀疏特征向量集，以及对所述物品稀疏特征信息集中的每个物品稀疏特征信息进行信息编码，以生成物品稀疏特征向量，得到物品稀疏特征向量集；Encoding each user sparse feature information in the user sparse feature information set to generate a user sparse feature vector, thereby obtaining a user sparse feature vector set, and encoding each item sparse feature information in the item sparse feature information set to generate an item sparse feature vector, thereby obtaining an item sparse feature vector set;
    对所述历史图像编码向量序列中的各个历史图像编码向量进行视觉偏好调整，得到调整后历史图像编码向量序列；Performing visual preference adjustment on each historical image coding vector in the historical image coding vector sequence to obtain an adjusted historical image coding vector sequence;
    根据所述调整后历史图像编码向量序列、所述主图编码向量、所述用户稀疏特征向量集和所述物品稀疏特征向量集，利用多头注意力机制模型，生成待推送给所述目标用户的、所述目标推荐物品对应的推荐图像集，其中，所述推荐图像集为所述创意图像特征信息集对应创意图像集的图像子集。According to the adjusted historical image encoding vector sequence, the main image encoding vector, the user sparse feature vector set and the item sparse feature vector set, a multi-head attention mechanism model is used to generate a recommended image set corresponding to the target recommended item to be pushed to the target user, wherein the recommended image set is an image subset of the creative image set corresponding to the creative image feature information set.
  2. 根据权利要求1所述的方法，其中，所述对所述历史图像编码向量序列中的各个历史图像编码向量进行视觉偏好调整，得到调整后历史图像编码向量序列，包括：The method according to claim 1, wherein the step of performing visual preference adjustment on each historical image coding vector in the historical image coding vector sequence to obtain an adjusted historical image coding vector sequence comprises:
    确定所述历史图像编码向量序列对应的视觉偏好信息，作为目标视觉偏好信息；Determining visual preference information corresponding to the historical image coding vector sequence as target visual preference information;
    根据所述目标视觉偏好信息，对所述历史图像编码向量序列中的各个历史图像编码向量进行调整，得到调整后历史图像编码向量序列。 According to the target visual preference information, each historical image coding vector in the historical image coding vector sequence is adjusted to obtain an adjusted historical image coding vector sequence.
  3. 根据权利要求1或2所述的方法，其中，所述根据所述调整后历史图像编码向量序列、所述主图编码向量、所述用户稀疏特征向量集和所述物品稀疏特征向量集，利用多头注意力机制模型，生成待推送给所述目标用户的、所述目标推荐物品对应的推荐图像集，包括：The method according to claim 1 or 2, wherein the step of generating a recommended image set corresponding to the target recommended item to be pushed to the target user using a multi-head attention mechanism model based on the adjusted historical image encoding vector sequence, the main image encoding vector, the user sparse feature vector set, and the item sparse feature vector set comprises:
    将所述用户稀疏特征向量集和所述物品稀疏特征向量集进行向量拼接，得到拼接稀疏特征向量；Performing vector concatenation on the user sparse feature vector set and the item sparse feature vector set to obtain a concatenated sparse feature vector;
    将所述调整后历史图像编码向量序列、所述主图编码向量、所述拼接稀疏特征向量输入至所述多头注意力机制模型，得到特征信息融合向量；Input the adjusted historical image encoding vector sequence, the main image encoding vector, and the spliced sparse feature vector into the multi-head attention mechanism model to obtain a feature information fusion vector;
    将所述拼接稀疏特征向量输入至全连接模型，得到输出向量；Inputting the concatenated sparse feature vector into a fully connected model to obtain an output vector;
    将所述特征信息融合向量和所述输出向量进行拼接，得到拼接向量；Splicing the feature information fusion vector and the output vector to obtain a spliced vector;
    利用预设损失函数，生成所述推荐图像集。The recommended image set is generated using a preset loss function.
  4. 根据权利要求1-3之一所述的方法，其中，所述对所述历史浏览创意图像序列中的每个历史浏览创意图像进行图编码处理，以生成历史图像编码向量，包括：The method according to any one of claims 1 to 3, wherein the step of performing image coding processing on each historically browsed creative image in the historically browsed creative image sequence to generate a historical image coding vector comprises:
    将所述历史浏览创意图像输入至预先训练的图编码模型，以生成历史图像编码向量。The historical browsed creative images are input into a pre-trained graph coding model to generate a historical image coding vector.
  5. 根据权利要求4所述的方法，其中，所述图编码模型包括：残差网络模型和多个全连接层；以及The method according to claim 4, wherein the graph coding model comprises: a residual network model and a plurality of fully connected layers; and
    所述将所述历史浏览创意图像输入至预先训练的图编码模型，以生成历史图像编码向量，包括：The step of inputting the historical browsing creative image into a pre-trained graph coding model to generate a historical image coding vector comprises:
    将所述历史浏览创意图像输入至所述残差网络模型，得到模型输出结果；Inputting the historical browsing creative image into the residual network model to obtain a model output result;
    将所述模型输出结果输入至所述多个全连接层，得到所述历史图像编码向量。The model output result is input into the multiple fully connected layers to obtain the historical image encoding vector.
  6. 根据权利要求3所述的方法，其中，所述利用预设损失函数， 生成所述推荐图像集，包括：The method according to claim 3, wherein the using a preset loss function, Generating the recommended image set includes:
    利用所述预设损失函数，生成针对所述拼接向量的创意图像分数集；Using the preset loss function, generating a creative image score set for the splicing vector;
    根据所述调整后历史图像编码向量序列、所述主图编码向量、所述用户稀疏特征向量集和所述物品稀疏特征向量集，利用多头注意力机制模型，生成所述目标推荐物品对应的推荐物品分数；Generate a recommended item score corresponding to the target recommended item using a multi-head attention mechanism model according to the adjusted historical image encoding vector sequence, the main image encoding vector, the user sparse feature vector set and the item sparse feature vector set;
    响应于确定所述推荐物品分数大于预定推荐物品数值，确定所述创意图像分数集中分数值大于预定推荐创意图像数值的创意图像分数，得到创意图像分数子集；In response to determining that the recommended item score is greater than a predetermined recommended item value, determining a creative image score having a score value greater than a predetermined recommended creative image value in the creative image score set, and obtaining a creative image score subset;
    将所述创意图像分数子集对应的创意图像集确定为所述推荐图像集。The creative image set corresponding to the creative image score subset is determined as the recommended image set.
  7. 根据权利要求6所述的方法，其中，所述方法还包括：The method according to claim 6, wherein the method further comprises:
    根据所述调整后历史图像编码向量序列、所述主图编码向量、所述用户稀疏特征向量集和所述物品稀疏特征向量集，利用多头注意力机制模型，生成所述目标推荐物品对应的推荐主图像分数；Generate a recommended main image score corresponding to the target recommended item using a multi-head attention mechanism model according to the adjusted historical image encoding vector sequence, the main image encoding vector, the user sparse feature vector set and the item sparse feature vector set;
    响应于确定所述推荐物品分数大于预定推荐物品数值、且所述推荐主图像分数大于预定推荐主图像数值，将所述主图像推送至所述目标用户对应的终端。In response to determining that the recommended item score is greater than a predetermined recommended item value and the recommended main image score is greater than a predetermined recommended main image value, the main image is pushed to a terminal corresponding to the target user.
  8. 一种推荐信息生成装置，包括：A device for generating recommendation information, comprising:
    获取单元，被配置成获取目标用户的历史浏览创意图像序列、所述目标用户的用户稀疏特征信息集、目标推荐物品的主图像和所述目标推荐物品的物品稀疏特征信息集，其中，所述物品稀疏特征信息集包括针对所述目标推荐物品的创意图像特征信息集；an acquisition unit, configured to acquire a target user's historical browsing creative image sequence, a user sparse feature information set of the target user, a main image of a target recommended item, and an item sparse feature information set of the target recommended item, wherein the item sparse feature information set includes a creative image feature information set for the target recommended item;
    图编码单元，被配置成对所述主图像进行图编码处理，得到主图编码向量，以及对所述历史浏览创意图像序列中的每个历史浏览创意图像进行图编码处理，以生成历史图像编码向量，得到历史图像编码向量序列；A picture coding unit is configured to perform picture coding processing on the main image to obtain a main picture coding vector, and to perform picture coding processing on each historical browsing creative image in the historical browsing creative image sequence to generate a historical image coding vector to obtain a historical image coding vector sequence;
    信息编码单元，被配置成对所述用户稀疏特征信息集中的每个用 户稀疏特征信息进行信息编码，以生成用户稀疏特征向量，得到用户稀疏特征向量集，以及对所述物品稀疏特征信息集中的每个物品稀疏特征信息进行信息编码，以生成物品稀疏特征向量，得到物品稀疏特征向量集；The information encoding unit is configured to encode each user in the user sparse feature information set. encoding the user sparse feature information to generate a user sparse feature vector to obtain a user sparse feature vector set, and encoding each item sparse feature information in the item sparse feature information set to generate an item sparse feature vector to obtain an item sparse feature vector set;
    调整单元，被配置成对所述历史图像编码向量序列中的各个历史图像编码向量进行视觉偏好调整，得到调整后历史图像编码向量序列；an adjustment unit configured to perform visual preference adjustment on each historical image coding vector in the historical image coding vector sequence to obtain an adjusted historical image coding vector sequence;
    生成单元，被配置成根据所述调整后历史图像编码向量序列、所述主图编码向量、所述用户稀疏特征向量集和所述物品稀疏特征向量集，利用多头注意力机制模型，生成待推送给所述目标用户的、所述目标推荐物品对应的推荐图像集。The generation unit is configured to generate a recommended image set corresponding to the target recommended item to be pushed to the target user by using a multi-head attention mechanism model based on the adjusted historical image encoding vector sequence, the main image encoding vector, the user sparse feature vector set and the item sparse feature vector set.
  9. 一种电子设备，包括：An electronic device, comprising:
    至少一个处理器；at least one processor;
    存储装置，其上存储有至少一个程序，a storage device having at least one program stored thereon,
    当所述至少一个程序被所述至少一个处理器执行，使得所述至少一个处理器实现如权利要求1-7中任一所述的方法。When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 7.
  10. 一种计算机可读介质，其上存储有计算机程序，其中，所述程序被处理器执行时实现如权利要求1-7中任一所述的方法。A computer readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.
  11. 一种计算机程序产品，包括计算机程序，所述计算机程序在被处理器执行时实现如权利要求1-7中任一所述的方法。 A computer program product comprises a computer program, wherein when the computer program is executed by a processor, the computer program implements the method according to any one of claims 1 to 7.
PCT/CN2023/121495 2022-09-28 2023-09-26 Recommendation-information generation method and apparatus, and device, medium and program product WO2024067568A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211194196.3A CN115757933A (en) 2022-09-28 2022-09-28 Recommendation information generation method, device, equipment, medium and program product
CN202211194196.3 2022-09-28

Publications (1)

Publication Number Publication Date
WO2024067568A1 true WO2024067568A1 (en) 2024-04-04

Family

ID=85350537

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/121495 WO2024067568A1 (en) 2022-09-28 2023-09-26 Recommendation-information generation method and apparatus, and device, medium and program product

Country Status (2)

Country Link
CN (1) CN115757933A (en)
WO (1) WO2024067568A1 (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115757933A (en) * 2022-09-28 2023-03-07 北京沃东天骏信息技术有限公司 Recommendation information generation method, device, equipment, medium and program product

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111144933A (en) * 2019-12-13 2020-05-12 北京三快在线科技有限公司 Commodity recommendation method and device, electronic equipment and storage medium
CN113377914A (en) * 2021-06-10 2021-09-10 北京沃东天骏信息技术有限公司 Recommended text generation method and device, electronic equipment and computer readable medium
CN113807926A (en) * 2021-09-26 2021-12-17 北京沃东天骏信息技术有限公司 Recommendation information generation method and device, electronic equipment and computer readable medium
CN115757933A (en) * 2022-09-28 2023-03-07 北京沃东天骏信息技术有限公司 Recommendation information generation method, device, equipment, medium and program product

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111144933A (en) * 2019-12-13 2020-05-12 北京三快在线科技有限公司 Commodity recommendation method and device, electronic equipment and storage medium
CN113377914A (en) * 2021-06-10 2021-09-10 北京沃东天骏信息技术有限公司 Recommended text generation method and device, electronic equipment and computer readable medium
CN113807926A (en) * 2021-09-26 2021-12-17 北京沃东天骏信息技术有限公司 Recommendation information generation method and device, electronic equipment and computer readable medium
CN115757933A (en) * 2022-09-28 2023-03-07 北京沃东天骏信息技术有限公司 Recommendation information generation method, device, equipment, medium and program product

Also Published As

Publication number Publication date
CN115757933A (en) 2023-03-07

Similar Documents

Publication Publication Date Title
WO2021196903A1 (en) Video processing method and device, readable medium and electronic device
WO2024067568A1 (en) Recommendation-information generation method and apparatus, and device, medium and program product
US20230326248A1 (en) Expression transformation method and apparatus, electronic device, and computer readable medium
WO2022105779A1 (en) Image processing method, model training method, and apparatus, medium, and device
CN111625645B (en) Training method and device for text generation model and electronic equipment
US20220385739A1 (en) Method and apparatus for generating prediction information, electronic device, and computer readable medium
US20230239546A1 (en) Theme video generation method and apparatus, electronic device, and readable storage medium
CN112785002A (en) Model construction optimization method, device, medium, and computer program product
CN114781351B (en) Bid file analysis method, device, equipment and medium based on power data
CN111967397A (en) Face image processing method and device, storage medium and electronic equipment
CN114881247A (en) Longitudinal federal feature derivation method, device and medium based on privacy computation
JP2024509077A (en) Information generation method, device, electronic device and computer readable medium
WO2023241650A1 (en) Push processing method and apparatus, and device and medium
WO2023221661A1 (en) Generation method and apparatus for user preferred item information, and electronic device and medium
WO2023179308A1 (en) Image description generation method and apparatus, device, medium, and product
CN115062119B (en) Government affair event handling recommendation method and device
CN116823984A (en) Element layout information generation method, device, apparatus, medium, and program product
CN114125485B (en) Image processing method, device, equipment and medium
US20230418794A1 (en) Data processing method, and non-transitory medium and electronic device
US20240134892A1 (en) Information generation method, apparatus, electronic device, and computer -readable medium
CN117009999B (en) Smart park data storage method, device, equipment and computer readable medium
WO2024074118A1 (en) Image processing method and apparatus, and device and storage medium
CN116049904B (en) Electronic bill encryption signature method based on blockchain multi-point certificate and electronic equipment
CN116800834B (en) Virtual gift merging method, device, electronic equipment and computer readable medium
CN115952315B (en) Campus monitoring video storage method, device, equipment, medium and program product