CN117893280A - Object list recommendation method and device - Google Patents
Object list recommendation method and device Download PDFInfo
- Publication number
- CN117893280A CN117893280A CN202311864492.4A CN202311864492A CN117893280A CN 117893280 A CN117893280 A CN 117893280A CN 202311864492 A CN202311864492 A CN 202311864492A CN 117893280 A CN117893280 A CN 117893280A
- Authority
- CN
- China
- Prior art keywords
- historical
- data
- feature vector
- training
- candidate object
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 45
- 239000013598 vector Substances 0.000 claims abstract description 496
- 238000010606 normalization Methods 0.000 claims abstract description 83
- 238000012545 processing Methods 0.000 claims abstract description 74
- 230000004913 activation Effects 0.000 claims abstract description 24
- 230000006399 behavior Effects 0.000 claims description 272
- 238000012549 training Methods 0.000 claims description 217
- 230000007613 environmental effect Effects 0.000 claims description 18
- 238000004590 computer program Methods 0.000 claims description 17
- 230000004927 fusion Effects 0.000 claims description 16
- 238000007526 fusion splicing Methods 0.000 claims description 13
- 238000000605 extraction Methods 0.000 claims description 11
- 125000004122 cyclic group Chemical group 0.000 claims description 5
- 230000004580 weight loss Effects 0.000 claims 1
- 230000006870 function Effects 0.000 description 24
- 238000004364 calculation method Methods 0.000 description 20
- 230000003213 activating effect Effects 0.000 description 18
- 230000003993 interaction Effects 0.000 description 13
- 238000010586 diagram Methods 0.000 description 9
- 230000008569 process Effects 0.000 description 8
- 238000004891 communication Methods 0.000 description 5
- 238000013507 mapping Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 4
- 230000010354 integration Effects 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 230000002093 peripheral effect Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000000513 principal component analysis Methods 0.000 description 1
- 238000012847 principal component analysis method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The disclosure relates to the technical field of data processing, and provides an object list recommending method and device. The method comprises the following steps: performing weight distribution on the historical behavior feature vector and at least one candidate object historical feature vector to obtain a weighting coefficient, performing vector normalization on the weighting coefficient, and determining a related normalized vector; the historical behavior feature vectors and the related normalization vectors are fused and spliced, and the fused cross feature vectors are subjected to weighted splicing to obtain object interest feature vectors; and performing activation weighting according to the object interest feature vectors, determining a first target object list recommended to the user, and increasing the dimensionality of the feature vectors, so that the richness of the feature information is improved, the click rate of candidate objects is improved through weight sorting, and the robustness of the model to the expression of the object feature information is improved.
Description
Technical Field
The disclosure relates to the technical field of data processing, and in particular relates to an object list recommending method and device.
Background
The recommendation system provides commodity information and suggestions for clients by using an e-commerce website, helps users determine what products should be purchased, and simulates sales staff to help clients complete the purchasing process. The personalized recommendation is to recommend information and commodities interested by a user to the user according to the interest characteristics and purchasing behavior of the user, the personalized recommendation system is an advanced business intelligent platform based on mass data mining to help an e-commerce website to provide complete personalized decision support and information service for shopping of customers of the electronic commerce website, but the current recommendation system only recommends historical click commodities to the user, ignores possible demands of the user, and carries out chain recommendation on the user according to interaction information of the commodity and single commodity through analysis of commodities interacted by the user in the past.
Therefore, the problem that the function efficiency of the recommended article is low, the effect is poor and the potential demands of the user part are ignored due to the single recommended information in the prior art.
Disclosure of Invention
In view of the above, the embodiments of the present disclosure provide a method and an apparatus for recommending an object list, so as to solve the problems in the prior art that the recommended article has low functional efficiency and poor effect due to the singleization of the recommended information, and neglect the potential demands of the user part.
In a first aspect of an embodiment of the present disclosure, there is provided an object list recommendation method, including: according to the historical behavior data of the object, acquiring the historical data of at least one candidate object associated with the historical behavior data of the object; extracting features of the historical behavior data of the object and the historical data of the candidate object to obtain a historical behavior feature vector corresponding to the historical behavior data of the object and a candidate object historical feature vector corresponding to the historical data of the candidate object; performing weight distribution on the historical behavior feature vector and at least one candidate object historical feature vector to obtain a weighting coefficient corresponding to the candidate object historical feature vector; carrying out vector normalization processing according to at least one candidate object historical feature vector and a weighting coefficient corresponding to the candidate object historical feature vector, and determining a relevant normalization vector corresponding to the candidate object historical data; fusion splicing is carried out on the historical behavior feature vector and the related normalized vector corresponding to the historical data of the candidate object, so that the cross feature vector corresponding to the historical behavior data of the object is obtained; performing weighted concatenation on the cross feature vector corresponding to the historical behavior data of the object and the candidate object attribute feature vector to obtain an object interest feature vector corresponding to the historical behavior data of the object; performing activation weighting according to object interest feature vectors corresponding to the historical behavior data of the object, and determining a first target weight sequence, wherein the first target weight sequence comprises weight values corresponding to the historical behavior data of the object and weight values corresponding to the historical data of the candidate object; and determining a first target object list recommended to the user according to the first target weight sequence, wherein the first target object list comprises objects and at least one candidate object.
In a second aspect of the embodiments of the present disclosure, there is provided an object list recommendation apparatus, including: the acquisition module is used for acquiring the historical data of at least one candidate object associated with the historical behavior data of the object according to the historical behavior data of the object; the first processing module is used for extracting features of the historical behavior data of the object and the historical data of the candidate object to obtain a historical behavior feature vector corresponding to the historical behavior data of the object and a candidate object historical feature vector corresponding to the historical data of the candidate object; the second processing module is used for carrying out weight distribution on the historical behavior feature vector and at least one candidate object historical feature vector to obtain a weighting coefficient corresponding to the candidate object historical feature vector; the third processing module is used for carrying out vector normalization processing according to at least one candidate object history feature vector and a weighting coefficient corresponding to the candidate object history feature vector, and determining a relevant normalization vector corresponding to the history data of the candidate object; the fourth processing module is used for obtaining a cross feature vector corresponding to the historical behavior data of the object by fusing and splicing the historical behavior feature vector and the related normalized vector corresponding to the historical data of the candidate object; the fifth processing module is used for carrying out weighted splicing on the cross feature vector corresponding to the historical behavior data of the object and the candidate object attribute feature vector to obtain an object interest feature vector corresponding to the historical behavior data of the object; the first determining module is used for performing activation weighting according to object interest feature vectors corresponding to the historical behavior data of the object, and determining a first target weight sequence, wherein the first target weight sequence comprises weight values corresponding to the historical behavior data of the object and weight values corresponding to the historical data of the candidate object; and the second determining module is used for determining a first target object list recommended to the user according to the first target weight sequence, wherein the first target object list comprises objects and at least one candidate object.
In a third aspect of the disclosed embodiments, an electronic device is provided, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the above method when executing the computer program.
In a fourth aspect of the disclosed embodiments, there is provided a readable storage medium storing a computer program which, when executed by a processor, implements the steps of the above method.
Compared with the prior art, the embodiment of the disclosure has the beneficial effects that: the method comprises the steps of obtaining historical data of at least one candidate object associated with historical behavior data of an object, carrying out feature extraction on the historical data to obtain a corresponding historical behavior feature vector and a candidate object historical feature vector, carrying out weight distribution on the historical behavior feature vector and the candidate object historical feature vector to obtain a weighting coefficient corresponding to each candidate object historical feature vector, normalizing the weighting coefficient and the corresponding candidate historical feature vector to obtain a relevant normalized vector corresponding to the historical data of the candidate object, carrying out fusion splicing on the relevant normalized vector and the historical behavior feature vector to obtain a cross feature vector, carrying out weighted splicing on the cross feature vector and the candidate object attribute feature vector to obtain an object interest feature vector, activating the object interest feature vector through an activating function, carrying out weighted processing on an activating result to obtain a first target weight sequence, and determining a first target object list according to the first target weight sequence, wherein the first target object list is used for displaying recommended objects and sequences of all objects to users, so that feature fusion dimensions are improved, information density of the feature vectors is improved, feature information is enriched, the weight of feature information is improved, the candidate object attribute information is improved, the candidate object interaction range is enlarged, and the candidate object interaction range is displayed.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings that are required for the embodiments or the description of the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present disclosure, and other drawings may be obtained according to these drawings without inventive effort for a person of ordinary skill in the art.
Fig. 1 is a scene schematic diagram of an application scene of an embodiment of the present disclosure;
fig. 2 is a flowchart of an object list recommendation method according to an embodiment of the present disclosure;
FIG. 3 is a schematic diagram of an object list recommendation model according to an embodiment of the present disclosure;
fig. 4 is a schematic structural diagram of an object list recommending apparatus according to an embodiment of the present disclosure;
fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure.
Detailed Description
In the following description, for purposes of explanation and not limitation, specific details are set forth, such as particular system configurations, techniques, etc. in order to provide a thorough understanding of the disclosed embodiments. However, it will be apparent to one skilled in the art that the present disclosure may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present disclosure with unnecessary detail.
It should be noted that, the user information (including, but not limited to, terminal device information, user personal information, etc.) and the data (including, but not limited to, data for presentation, analyzed data, etc.) related to the present disclosure are information and data authorized by the user or sufficiently authorized by each party.
An object list recommendation method and apparatus according to embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.
Fig. 1 is a scene diagram of an application scene of an embodiment of the present disclosure. The application scenario may include terminal devices 1,2 and 3, a server 4 and a network 5.
The terminal devices 1, 2 and 3 may be hardware or software. When the terminal devices 1, 2 and 3 are hardware, they may be various electronic devices having a display screen and supporting communication with the server 4, including but not limited to smart phones, tablet computers, laptop portable computers, desktop computers, and the like; when the terminal apparatuses 1, 2, and 3 are software, they can be installed in the electronic apparatus as above. The terminal devices 1, 2 and 3 may be implemented as a plurality of software or software modules, or as a single software or software module, to which the embodiments of the present disclosure are not limited. Further, various applications, such as a data processing application, an instant messaging tool, social platform software, a search class application, a shopping class application, and the like, may be installed on the terminal devices 1, 2, and 3.
The server 4 may be a server that provides various services, for example, a background server that receives a request transmitted from a terminal device with which communication connection is established, and the background server may perform processing such as receiving and analyzing the request transmitted from the terminal device and generate a processing result. The server 4 may be a server, a server cluster formed by a plurality of servers, or a cloud computing service center, which is not limited in the embodiment of the present disclosure.
The server 4 may be hardware or software. When the server 4 is hardware, it may be various electronic devices that provide various services to the terminal devices 1, 2, and 3. When the server 4 is software, it may be a plurality of software or software modules providing various services to the terminal devices 1, 2, and 3, or may be a single software or software module providing various services to the terminal devices 1, 2, and 3, which is not limited by the embodiments of the present disclosure.
The network 5 may be a wired network using coaxial cable, twisted pair wire, and optical fiber connection, or may be a wireless network that can implement interconnection of various Communication devices without wiring, for example, bluetooth (Bluetooth), near Field Communication (NFC), infrared (Infrared), etc., which are not limited by the embodiments of the present disclosure.
The user can establish a communication connection with the server 4 via the network 5 through the terminal devices 1,2, and 3 to receive or transmit information or the like. Specifically, after the user imports the collected data of the interest points to the server 4, the server 4 may obtain the historical behavior data of the object and the historical data of at least one candidate object associated with the historical behavior data of the object from the terminal device 1,2 or 3, perform feature extraction on the foregoing data to obtain a corresponding historical behavior feature vector and a candidate object historical feature vector, perform weight distribution on the historical behavior feature vector and the candidate object historical feature vector to obtain a weighting coefficient corresponding to each candidate object historical feature vector, normalize the weighting coefficient and the corresponding candidate historical feature vector to obtain a correlation normalized vector corresponding to the historical data of the candidate object, perform fusion splicing on the correlation normalized vector and the historical behavior feature vector to obtain a cross feature vector, perform weighted splicing on the cross feature vector and the candidate object attribute feature vector to obtain an object interest feature vector, activate the object interest feature vector through an activation function, perform weighted processing on an activation result to obtain a first target weight sequence, and determine a first target weight sequence according to the first target weight sequence, where the first target weight sequence is used for displaying the recommendation sequence to the user and all the objects.
It should be noted that the specific types, numbers and combinations of the terminal devices 1,2 and 3, the server 4 and the network 5 may be adjusted according to the actual requirements of the application scenario, which is not limited by the embodiment of the present disclosure.
Fig. 2 is a flowchart of an object list recommendation method according to an embodiment of the present disclosure. The object list recommendation method of fig. 2 may be performed by the server of fig. 1. As shown in fig. 2, the object list recommendation method includes:
Step 201, according to the historical behavior data of the object, obtaining the historical data of at least one candidate object associated with the historical behavior data of the object.
Specifically, the object may be an object used for having interaction with the user in the recommendation system, including but not limited to merchandise, video or pictures, wherein the interaction may be collection, forwarding, or purchasing, and the candidate object may be, but is not limited to, merchandise, video or pictures, etc. recommended to the user by the system but not having interaction with the user, and the total number of the object and the candidate object that can be recommended to the user at a time is different according to the type of the terminal device.
The historical behavior data may be related data of the object selected by the user in the recommendation system in a historical manner, and is used for representing characteristic information of the selected object in a historical manner, including but not limited to historical behavior records of the object, historical state data of the object, historical track data of the object, and the like, and is not limited herein.
The historical data of the candidate object may be feature information corresponding to other objects which are not selected by the user and are recommended together with the object selected by the user in the recommendation system, including but not limited to the historical behavior record of the candidate object, the historical state data of the candidate object, the historical track data of the candidate object, or the like, and is not limited herein, wherein the historical data of the candidate object and the historical behavior data of the object associated with the candidate object are in the same recommendation list.
By acquiring the historical behavior data of the object, the historical data of at least one candidate object in the same historical recommendation list with the historical behavior data of the object can be obtained, for example, in the historical recommendation list, the object A and the candidate object B, C, D are included, wherein the object A is an object with interactive behavior with a user, and the candidate object B, C, D is an object which is not operated by the user, so that the historical data of the candidate object is associated with the historical behavior data of the object, the historical data of the obtained candidate object is increased, the richness of data recommendation is improved, and the selection probability of the candidate object is improved.
Step 202, extracting features from the historical behavior data of the object and the historical data of the candidate object to obtain a historical behavior feature vector corresponding to the historical behavior data of the object and a candidate object historical feature vector corresponding to the historical data of the candidate object.
Specifically, the historical behavior feature vector may be a feature vector corresponding to the historical behavior data of the object, which is used for reducing the dimension of the historical behavior data of the object into a vector, so as to facilitate the recognition and operation of the model.
The candidate object history feature vector can be a feature vector corresponding to the history data of the candidate object, and is used for reducing the dimension of the history data of the candidate object into a vector, so that the recognition and the operation of the model are facilitated.
The corresponding historical behavior feature vector and the corresponding candidate object historical feature vector are obtained by carrying out feature extraction on the historical behavior data of the object and the historical data of the candidate object, and the historical behavior data of the object and the historical data of the candidate object can be subjected to feature extraction through principal component analysis, linear discriminant analysis, local binary pattern and the like, for example, the principal component analysis method is used for projecting the historical behavior data of the object and the historical data of the candidate object to the main direction of the data through a covariance matrix, so that the dimension of the data is reduced, various feature information is input into a follow-up object list recommendation model in a vector form for processing, the processing amount is reduced, and the processing flow is simplified.
And 203, performing weight distribution on the historical behavior feature vector and at least one candidate object historical feature vector to obtain a weighting coefficient corresponding to the candidate object historical feature vector.
Specifically, the weighting coefficient corresponding to the candidate object history feature vector may be a total weighting coefficient, where the total weighting coefficient may be obtained by performing exponential normalization processing on the individual weighting coefficient corresponding to each different candidate object history feature vector, and the individual weighting coefficient corresponding to each different candidate object history feature vector may be obtained by an expression, for example: α ij=Wc*Ei*Ws*Eij+Ej, where W c and W s are two learnable parameter matrices respectively, E i is a historical behavior feature vector, E ij represents a j candidate object historical feature vector around the historical behavior data of the object, and E j is a one-dimensional vector representing a distance between the historical data of the candidate object and the historical behavior data of the object.
And 204, carrying out vector normalization processing according to at least one candidate object history feature vector and the weighting coefficient corresponding to the candidate object history feature vector, and determining the relevant normalized vector corresponding to the history data of the candidate object.
Specifically, the overall weighting coefficient is obtained by performing exponential normalization processing on the individual weighting coefficient corresponding to each different candidate object history feature vector, and the overall weighting coefficient can be obtained by an expression, for example: Wherein alpha ij is an independent weighting coefficient corresponding to each different candidate object history feature vector, mu ij is a weighting coefficient corresponding to the candidate object history feature vector, and the weighting coefficient is obtained through the attention mechanism calculation of the expression, so that data support is provided for the relevant normalized vector corresponding to the history data of the candidate object to be calculated later, the subsequent processing flow is simplified, and the processing efficiency of the object list recommendation model is improved.
The relevant normalized vector corresponding to the historical data of the candidate object may be a normalized vector capable of representing features of the historical data of all candidate objects corresponding to the historical behavior data of the object, and may be obtained by an expression, for example: The method comprises the steps that the whole peripheral information of the historical behavior data i of the object, namely, the relevant normalized vector corresponding to the historical data of the candidate object is marked as v i, M candidate objects are shared around the historical behavior data i of the object, e dij represents the j candidate object historical feature vector, W V is a parameter matrix to be learned, relevant normalized vectors corresponding to the historical data of all candidate objects are obtained through the vector normalization processing, so that feature information of the historical data of all candidate objects is unified, the calculation process is simplified, the processing efficiency of an object list recommendation model is improved, the click rate of the candidate objects in a recommendation list is improved, and the accuracy of the object list recommendation model is improved.
And 205, fusing and splicing the historical behavior feature vector and the related normalized vector corresponding to the historical data of the candidate object to obtain the cross feature vector corresponding to the historical behavior data of the object.
Specifically, the cross feature vector corresponding to the historical behavior data of the object may be used to indicate fused cross feature information of the historical behavior data of the object and the historical data of the candidate object.
Through series fusion, a plurality of features with different characteristics are fused to form a new and more comprehensive feature, different features are connected in sequence to form a longer feature vector, after the hidden layer processing of the neural network, the representation vectors with different input features can be spliced to obtain richer feature representation, the feature vector can contain more information, and the cross feature vector corresponding to the historical behavior data of the object can be obtained through an expression, for example: i i=concat(xci,vi,xci-vi,xci*vi), wherein x ci is a historical behavior feature vector, v i is a candidate object historical feature vector, x ci-vi is a difference between the two, and x ci*vi is a product of the two, so that fusion characterization is enriched, a nonlinear relation is increased, feature expression of the cross feature vector is improved through multidimensional interaction of the two vectors, and information understanding of prediction probability of historical behavior data of an object and corresponding historical data of at least one candidate object is enhanced.
And 206, carrying out weighted stitching on the cross feature vector corresponding to the historical behavior data of the object and the candidate object attribute feature vector to obtain the object interest feature vector corresponding to the historical behavior data of the object.
Specifically, the object interest feature vector corresponding to the historical behavior data of the object may be used to indicate the interest expression of the user of the object list recommendation model, and the object interest feature includes, but is not limited to, a sports interest, a science and technology interest, or an artistic interest, etc., and may be obtained by calculating the relationship between the cross feature vector corresponding to the historical behavior data of the object and the candidate object attribute feature vector through a self-attention mechanism.
The candidate attribute feature vector may be a vector expressing candidate attribute feature information including, but not limited to, basic information, quality information, price information, or technical parameters of the candidate.
The object interest feature vector corresponding to the historical behavior data of the object can be obtained by an expression, for example: Wherein I i is the cross feature vector corresponding to the ith historical behavior data of the object, e t is the attribute feature vector of the candidate object, I j is the cross feature vector corresponding to the jth historical behavior data of the object, N is the total number of the cross feature vectors corresponding to the historical behavior data of the object, and U is the object interest feature vector corresponding to the historical behavior data of the object, so that the historical data of the candidate object and the historical behavior data of the object are fused, the recommendation times of the candidate object are improved, the robustness of the object recommendation list is improved, the integrity of the object recommendation list is ensured, and the application range of the object list recommendation model is enlarged.
Step 207, performing activation weighting according to the object interest feature vector corresponding to the historical behavior data of the object, and determining a first target weight sequence, where the first target weight sequence includes a weight value corresponding to the historical behavior data of the object and a weight value corresponding to the historical data of the candidate object.
Specifically, the first target weight sequence may be a sequence of weight values corresponding to historical behavior data of the objects and weight values corresponding to historical data of each candidate object, and is used for recommending to a user of the recommendation system, and determining the sequence of recommendation according to the sequence of the weight values, so that the predicted interaction probability of each object can be represented through the corresponding weight value of each object, the recommendation times of the candidate objects are improved, the robustness of the object recommendation list is improved, and the integrity of the object recommendation list is ensured.
For example, as one example, when a user clicks on an application of an e-commerce platform, historical usage data of the user on the e-commerce platform is obtained, including: and clicking the commodity A, calculating and processing the candidate commodities B, C, D and E displayed on the same screen through the object list recommendation model to obtain a first target weight sequence corresponding to the user, wherein the commodity A weight value is 0.5, the commodity B weight value is 0.2, the commodity C weight value is 0.03, the commodity D weight value is 0.06, and the commodity E weight value is 0.21, and the sequence of the first target weight sequence corresponding to the login information of the user in the recommendation system is 0.5, 0.21, 0.2, 0.06 and 0.03.
Step 208, determining a first target object list recommended to the user according to the first target weight sequence, wherein the first target object list contains objects and at least one candidate object.
In particular, a user may be a person using a product or service, and in the internet and e-commerce arts, a user generally refers to a person accessing a website, using an application, or purchasing a product or service, including but not limited to a consumer or merchant.
The first target object list may be an object or a candidate object set corresponding to each weight value in the first target weight sequence, and is used for indicating a recommendation sequence of the object and each candidate object when the object is recommended to the user, for example, when the user points out an application of the e-commerce platform, historical usage data of the user on the e-commerce platform is obtained, which includes: the method comprises the steps that a commodity A is clicked, candidate commodities B, C, D and E displayed on the same screen are calculated and processed through an object list recommendation model, a first target weight sequence corresponding to login information of a user in a recommendation system is obtained, wherein the commodity A weight value is 0.5, the commodity B weight value is 0.2, the commodity C weight value is 0.03, the commodity D weight value is 0.06, the commodity E weight value is 0.21, the sequence of the first target weight sequence corresponding to the login information of the user in the recommendation system is 0.5, 0.21, 0.2, 0.06 and 0.03, and the corresponding first target object list recommendation objects and the sequence thereof comprise: the weight values obtained through calculation and processing are used for recommending an object list to a user of a recommendation system, the recommending information and characteristics are increased, the accuracy of an object list recommending model is improved, the robustness of the model is improved, and the probability of selecting candidate objects is improved.
According to the technical scheme provided by the embodiment of the disclosure, by acquiring the historical data of at least one candidate object associated with the historical behavior data of the object, performing feature extraction on the data to obtain a corresponding historical behavior feature vector and a candidate object historical feature vector, performing weight distribution on the historical behavior feature vector and the candidate object historical feature vector to obtain a weighting coefficient corresponding to each candidate object historical feature vector, normalizing the weighting coefficient and the corresponding candidate historical feature vector to obtain a related normalized vector corresponding to the historical data of the candidate object, performing fusion splicing on the related normalized vector and the historical behavior feature vector to obtain a cross feature vector, performing weighted splicing on the cross feature vector and the candidate object attribute feature vector to obtain an object interest feature vector, activating the object interest feature vector through an activating function, performing weighted processing on an activating result to obtain a first target weight sequence, and determining a first target weight sequence according to the first target weight sequence.
In some embodiments, before performing activation weighting according to the object interest feature vector corresponding to the historical behavior data of the object, determining the first target weight sequence further includes: acquiring historical environment information data, wherein the historical environment information data is used for indicating historical environment information of terminal equipment corresponding to the object and the candidate object; extracting features of the historical environmental information data to obtain environmental information feature vectors corresponding to the historical environmental information data; activating and weighting according to object interest feature vectors corresponding to the historical behavior data of the object and environment information feature vectors corresponding to the historical environment information data, and determining a second target weight sequence, wherein the second target weight sequence comprises weight values corresponding to the historical behavior data of the object and weight values corresponding to the historical data of the candidate object; and determining a second target object list recommended to the user according to the second target weight sequence, wherein the ordering of the objects and the candidate objects in the second target object list and the ordering of the objects and the candidate objects in the first target object list are the same or different.
Specifically, the historical environment information data may be data of historical environment information of the terminal device used by the presentation object and the candidate object, where the historical environment information includes, but is not limited to, historical geographic location information, historical network environment information, historical device status information, or historical application usage information.
The environment information feature vector can be a vector obtained by feature extraction and dimension reduction of the historical environment information data, the environment information feature vector contains feature information of the historical environment information data, and after dimension reduction, the model is convenient to process and calculate, and the calculated amount and the processing flow are simplified.
The second target weight sequence can be a target weight sequence considering the historical environment information data on the basis of the first target weight sequence, so that the characteristics of the historical environment information data are increased, and the fusion characterization is enriched.
The second target object list may be an object or a candidate object set corresponding to each weight value in the second target weight sequence, and is used for indicating a recommendation sequence of the object and each candidate object when the object is recommended to the user, where the order of the object and the candidate object in the second target object list and the order of the object and the candidate object in the first target object list are the same or different, for example, when the order of the object and the candidate object in the second target object list and the order of the object and the candidate object in the first target object list are the same, as an example, the order of the second target weight sequence corresponding to the information of the user in the recommendation system is 0.45, 0.18 for the commodity A weight value, 0.06 for the commodity B weight value, 0.12 for the commodity D weight value, 0.19 for the commodity E weight value, and the order of the second target weight sequence corresponding to the information of the user in the recommendation system is 0.45, 0.19, 0.18, 0.06 for the commodity E, and the order corresponding to the information of the user in the recommendation system is obtained by calculation and processing of the object list recommendation model: commodity a, commodity E, commodity B, commodity D, commodity C.
For example, when the orders of the objects in the second target object list and the candidate objects in the candidate object list are different from those in the first target object list, as an example, the clicked commodity a, the candidate commodities B, C, D and E displayed on the same screen, through calculation and processing of the object list recommendation model, a second target weight sequence corresponding to the login information of the user in the recommendation system is obtained, where the commodity a weight value is 0.35, the commodity B weight value is 0.2, the commodity C weight value is 0.16, the commodity D weight value is 0.15, and the commodity E weight value is 0.14, the order of the second target weight sequence corresponding to the login information of the user in the recommendation system is 0.35, 0.2, 0.16, 0.15 and 0.14, and the corresponding second target object list recommendation objects include: commodity a, commodity B, commodity C, commodity D, commodity E.
The activating and weighting of the object interest feature vector corresponding to the historical behavior data of the object and the environment information feature vector corresponding to the historical environment information data may specifically be that the object interest feature vector and the environment information feature vector are normalized through an activating function, and mapped to a mapping interval of 0-1, wherein the mapping value is a weight value corresponding to the object and each candidate object, and the activating function includes, but is not limited to, a softmax activating function, a sigmoid activating function, a ReLU function, a Tanh function, or the like.
According to the technical scheme provided by the embodiment of the disclosure, by acquiring the historical environment information data, extracting the characteristics of the historical environment information data, carrying out activation weighting on the extracted environment information characteristic vector and the object interest characteristic vector to obtain the second target weight sequence, sorting according to the magnitude of the weight value in the second target weight sequence to obtain the corresponding second target object list, and displaying and recommending the objects and the candidate objects contained in the list to the user of the recommendation system according to the magnitude sequence of the weight value, thereby improving the characteristic richness of the objects and the candidate objects, increasing the quantity and the category of the object interest characteristic, improving the accuracy of the object list recommendation model, improving the robustness of the object list recommendation model, and expanding the application range of the object list recommendation model.
In some embodiments, vector normalization processing is performed according to at least one candidate object history feature vector and a weighting coefficient corresponding to the candidate object history feature vector, and determining a relevant normalized vector corresponding to the candidate object history feature vector includes: according to the at least one candidate object history feature vector, carrying out exponential normalization processing on the weighting coefficient corresponding to each candidate object history feature vector to obtain a normalization result corresponding to each candidate object history feature vector; adding the normalization result corresponding to each candidate object history feature vector to obtain a normalization addition result; dividing the normalization result and the normalization summation result corresponding to each candidate object history feature vector to obtain a weighting coefficient corresponding to each candidate object history feature vector; and determining a relevant normalized vector corresponding to the historical data of the candidate object according to the weighting coefficient corresponding to the historical feature vector of the candidate object and the historical feature vector of the candidate object.
Specifically, the weighting coefficient corresponding to each candidate object history feature vector is subjected to exponential normalization processing to obtain a normalization result corresponding to each candidate object history feature vector, for example, the normalization result may be expressed by the following expression: exp (alpha ij) is obtained, wherein alpha ij can be an independent weighting coefficient corresponding to each different candidate object history feature vector, and through exponential normalization, the over fitting of the model to the candidate object history feature vector is reduced, and the robustness of the object list recommendation model is enhanced.
And adding the normalization results corresponding to the historical feature vectors of each candidate object to obtain a normalization addition result, wherein the addition result can be obtained through expression calculation, for example: The normalized sum contains an exponential normalization result for each candidate historical feature vector.
The weighting coefficient corresponding to each candidate object history feature vector may be obtained by dividing the normalization result and the normalization summation result corresponding to each candidate object history feature vector, and the whole of the weighting coefficient may be obtained by expression calculation, for example: Wherein alpha ij is an independent weighting coefficient corresponding to the j-th candidate object history feature vector around the object history behavior data i, mu ij is a weighting coefficient corresponding to the candidate object history feature vector, and M candidate objects are shared around the object history behavior data i.
And then, through the total weighting coefficient corresponding to the candidate object history feature vector and combining with the corresponding candidate object history feature vector, obtaining a relevant normalized vector corresponding to the history data of the candidate object through integration calculation, wherein the relevant normalized vector can be obtained through expression calculation, for example: The method comprises the steps that the whole peripheral information of the historical behavior data i of an object, namely a relevant normalized vector corresponding to the historical data of the candidate object is marked as v i, M candidate objects are shared around the historical behavior data i of the object, e dij represents a j candidate object historical feature vector, W V is a parameter matrix to be learned, and the relevant normalized vector is finally determined.
According to the technical scheme provided by the embodiment of the disclosure, the weighting coefficient corresponding to each candidate object history feature vector is subjected to exponential normalization, the normalization result and the summation result are added, the weighting value of each candidate object history feature vector in the total normalization result summation is obtained by dividing the normalization result and the summation result, namely the weighting coefficient corresponding to each candidate object history feature vector, so that when the history data of the candidate object is introduced, the corresponding weighting coefficient is also obtained by calculation and used for indicating the weight of the history data of each candidate object, different weights of the history data of each candidate object are determined according to different feature information, the richness of recommendation information is improved, the recommendation degree of the candidate object is improved, and the robustness of the object list recommendation model is enhanced.
In some embodiments, the method for obtaining the cross feature vector corresponding to the historical behavior data of the object by fusion splicing the historical behavior feature vector and the related normalized vector corresponding to the historical data of the candidate object includes: subtracting the relevant normalized vector corresponding to the historical behavior feature vector and the historical data of the candidate object to obtain a difference result; multiplying the historical behavior feature vector and a related normalized vector corresponding to the historical data of the candidate object to obtain a product result; and according to the preset dimension, carrying out series fusion on the historical behavior feature vector, the related normalized vector corresponding to the historical data of the candidate object, the difference result and the product result to obtain the cross feature vector corresponding to the historical behavior data of the object.
Specifically, subtraction processing is performed on the historical behavior feature vector and a relevant normalized vector corresponding to the historical data of the candidate object, so as to obtain a difference result, where the difference result can be obtained by expression calculation, for example: x ci-vi, wherein x ci is a historical behavior feature vector, and v i is a relevant normalized vector corresponding to the historical data of the candidate object, so that vectors with different lengths are added.
Multiplying the historical behavior feature vector and the related normalized vector corresponding to the historical data of the candidate object to obtain a product result, wherein the product result can be obtained through expression calculation, for example: x ci*vi, wherein x ci is a historical behavior feature vector, and v i is a relevant normalized vector corresponding to the historical data of the candidate object, so that the richness of the vector length is improved.
The cross feature vector may be a feature vector of a preset dimension obtained by performing series fusion on the historical behavior feature vector, a related normalized vector corresponding to the historical data of the candidate object, a difference result, and a product result.
The preset dimension may be a target dimension preset according to an actual application scene, and is used for indicating a dimension of the cross feature vector.
According to the technical scheme provided by the embodiment of the disclosure, the historical behavior feature vector and the related normalization vector corresponding to the historical data of the candidate object are subtracted, the historical behavior feature vector and the related normalization vector corresponding to the historical data of the candidate object are multiplied, the twice processing result, the historical behavior feature vector and the related normalization vector corresponding to the historical data of the candidate object are input into the hidden layer of the neural network for series fusion, and the cross feature vector corresponding to the historical behavior data of the object with higher richness is obtained, so that the complexity and the expression capacity of the object list recommendation model are increased, and the prediction precision of the object list recommendation model is improved.
In some embodiments, performing weighted concatenation on a cross feature vector corresponding to historical behavior data of an object and a candidate object attribute feature vector to obtain an object interest feature vector corresponding to the historical behavior data of the object, where the weighted concatenation includes: multiplying the cross feature vector corresponding to the historical behavior data of the object with the candidate object attribute feature vector, and carrying out exponential normalization on the processing result to obtain an attribute normalization result; performing self-attention weight processing on the attribute normalization result to obtain an attribute self-attention weight result; and carrying out weighted summation on the attribute self-attention weight result to obtain the object interest feature vector corresponding to the historical behavior data of the object.
Specifically, the cross feature vector corresponding to the historical behavior data of the object and the candidate object attribute feature vector are multiplied, which can be obtained by expression calculation, for example: e t*Ii, and performing index normalization on the processing result to obtain an attribute normalization result, which can be obtained by calculating an expression, for example: exp (e t*Ii), where I i is a cross feature vector corresponding to the ith historical behavior data of the object, and e t is a candidate object attribute feature vector.
The attribute normalization result is subjected to self-attention weight processing, and the attribute self-attention weight result can be obtained through expression calculation, for example: Wherein I j is the cross feature vector corresponding to the jth historical behavior data of the object, and N is the total number of cross feature vectors corresponding to the historical behavior data of the object.
The weighted summation is performed on the attribute self-attention weight results to obtain object interest feature vectors corresponding to the historical behavior data of the object, and the object interest feature vectors can be obtained through expression calculation, for example: And U is an object interest feature vector corresponding to the historical behavior data of the object.
According to the technical scheme provided by the embodiment of the disclosure, through multiplying the cross feature vector corresponding to the historical behavior data of the object and the attribute feature vector of the candidate object, then carrying out index normalization processing, carrying out self-attention weighting processing on the attribute normalization result, and carrying out weighted summation on the processed weighting result to obtain the object interest feature vector corresponding to the historical behavior data of the object, thereby improving the recommendation times of the candidate object, improving the robustness of an object recommendation list, ensuring the integrity of the object recommendation list and expanding the application range of an object list recommendation model.
In some embodiments, before obtaining the historical data of the at least one candidate object associated with the historical behavior data of the object from the historical behavior data of the object, further comprising: acquiring historical behavior data of a training object, historical data of a training candidate object, labels corresponding to the historical data of the training candidate object, attribute data of the training candidate object and training historical environment information data, wherein the labels corresponding to the historical data of the training candidate object are used for indicating real weight values corresponding to the historical data of the training candidate object; inputting the historical behavior data of the training object, the historical data of the training candidate object, the attribute data of the training candidate object and the historical environmental information data of the training history to the object list recommendation model for feature extraction to obtain a historical behavior training vector corresponding to the historical behavior data of the training object, a candidate object historical training vector corresponding to the historical data of the training candidate object, a candidate object attribute training vector corresponding to the attribute data of the training candidate object and a historical environmental information training vector corresponding to the historical environmental information data of the training candidate object; performing weight distribution on the historical behavior training vector and at least one candidate object historical training vector to obtain a weighting coefficient corresponding to the candidate object historical training vector; carrying out vector normalization processing according to at least one candidate object historical training vector and a weighting coefficient corresponding to the candidate object historical training vector, and determining a relevant normalization vector corresponding to the training candidate object data; the cross feature vectors corresponding to the historical behavior data of the training object are obtained by fusion and splicing of the historical behavior training vectors and the related normalization vectors corresponding to the data of the training candidate object; performing weighted concatenation on the cross feature vector corresponding to the historical behavior data of the training object and the candidate object attribute training vector to obtain an object interest training vector corresponding to the historical behavior data of the training object; activating and weighting according to an object interest training vector corresponding to the historical behavior data of the training object and an environment information training vector, and determining a training target weight sequence, wherein the training target weight sequence comprises a weight value corresponding to the historical behavior data of the training object and a weight value corresponding to the historical data of the training candidate object, and determining a weighting loss according to the training target weight sequence and a label corresponding to the historical data of the training candidate object; and updating parameters in the object list recommendation model according to the weighting loss in a cyclic iteration mode.
Specifically, the historical behavior data of the training object may include, but is not limited to, a historical behavior record of the training object, historical state data of the training object, or historical track data of the training object, for example, as an example, the training object is the object a, the historical behavior record of a may be a historical interaction behavior record with the user, the historical state data may be an up-or-down state of a, the historical track data may be a collection or deletion track of a, and the like.
The historical data of the training candidates may include, but is not limited to, historical interaction data, historical state data, historical behavioral records, etc. of the training candidates, which may be understood in connection with the embodiments of the training candidates described above.
The tag corresponding to the history data of the training candidate may be used to indicate a true weight value of the history data of the training candidate, which may be 0.5, 0.2, or 0.06, etc., and is not limited herein.
The training candidate attribute data may be data indicative of training candidate attribute information including, but not limited to: basic information, quality information, price information, or technical parameters of the training candidate, for example, the training candidate is candidate B, the basic information of candidate B may be a name, the quality information may be superior, the price information may be 200 yuan, and the technical parameters may be that the length of B is 20cm.
The training history environment information data may be history environment information indicating the terminal device corresponding to the training object and the training candidate object, and may be geographical location information, network environment information, or device status information, which is not limited herein.
And extracting features of the different data to obtain a historical behavior training vector corresponding to the historical behavior data of the training object, a candidate object historical training vector corresponding to the historical data of the training candidate object, a candidate object attribute training vector corresponding to the attribute data of the training candidate object and a historical environment information training vector corresponding to the training environment information data.
According to the expression: alpha ij=Wc*Ei*Ws*Eij+Ej, wherein W c and W s are two learnable parameter matrixes respectively, E i is a historical behavior training vector, E ij represents a historical training vector corresponding to a j-th training candidate around the historical behavior data of the training object, E j is a one-dimensional vector, represents the distance between the historical data of the training candidate and the historical behavior data of the training object, and calculates a weighting coefficient corresponding to the historical training vector of the candidate object.
According to , alpha ij is an independent weighting coefficient corresponding to each different candidate object historical training vector, mu ij is a weighting coefficient corresponding to the candidate object historical training vector, the training total weighting coefficient is obtained through calculation through the attention mechanism of the expression, the training total weighting coefficient is brought into the expression/> , the whole peripheral information of the historical behavior data i of the training object, namely the relevant normalization vector corresponding to the historical data of the training candidate object is marked as v i, M training candidate objects are shared around the historical behavior data i of the training object, ed ij represents the j candidate object historical training vector, W V is a parameter matrix to be learned, and the relevant normalization vector corresponding to the data of the training candidate object is obtained through calculation.
And carrying out fusion splicing by bringing the relevant normalized vector corresponding to the historical behavior training vector and the data of the training candidate object into an expression I i=concat(xci,vi,xci-vi,xci*vi), wherein x ci is the historical behavior training vector, v i is the candidate object historical training vector, x ci-vi is the difference between the historical behavior training vector and the candidate object historical training vector, and x ci*vi is the product of the historical behavior training vector and the candidate object historical training vector, and calculating to obtain the cross feature vector corresponding to the historical behavior data of the training object.
And carrying the cross feature vector corresponding to the historical behavior data of the training object into an expression , wherein I i is the cross feature vector corresponding to the ith historical behavior data of the training object, e t is the candidate object attribute training vector, I j is the cross feature vector corresponding to the jth historical behavior data of the training object, N is the total number of the cross feature vectors corresponding to the historical behavior data of the training object, and thus calculating the object interest training vector U corresponding to the historical behavior data of the training object.
Normalizing the object interest training vector and the environment information training vector according to the object interest training vector and the environment information training vector corresponding to the historical behavior data of the training object through an activation function, mapping the object interest training vector and the environment information training vector to a mapping interval of 0-1, wherein the mapping value is a weight value corresponding to the object and each candidate object, and the activation function comprises, but is not limited to, a softmax activation function, a sigmoid activation function, a ReLU function, a Tanh function or the like, and performing activation weighting to determine a training target weight sequence.
And determining a weighting loss according to the training target weight sequence and the label corresponding to the historical data of the training candidate object, wherein the weighting loss can be obtained through a cross entropy loss calculation method, a mean square error loss calculation method, a root mean square error loss calculation method or the like, the method is not limited, and finally, parameters in the object list recommendation model are updated according to the weighting loss in a cyclic iteration mode.
According to the technical proposal provided by the embodiment of the disclosure, weight distribution is carried out on the historical behavior training vector and at least one candidate object historical training vector by acquiring the historical behavior data of the training object, the historical data of the training candidate object, the label corresponding to the historical data of the training candidate object, the attribute data of the training candidate object and the historical environment information data of the training history, and extracting the characteristics of the data to obtain the historical behavior training vector corresponding to the historical behavior data of the training object, the candidate object historical training vector corresponding to the historical data of the training candidate object, the candidate object attribute training vector corresponding to the attribute data of the training candidate object and the historical environment information training vector corresponding to the historical environment information data of the training candidate object, the vector normalization is carried out according to the weighting coefficient obtained by the weight distribution, obtaining a relevant normalized vector corresponding to the data of the training candidate object, carrying out fusion splicing on the relevant normalized vector and the historical behavior training vector to obtain a cross feature vector corresponding to the historical behavior data of the training object, carrying out weighted splicing on the cross feature vector and the candidate object attribute training vector to obtain an object interest training vector corresponding to the historical behavior data of the training object, carrying out activation weighting according to the object interest training vector corresponding to the historical behavior data of the training object and the environment information training vector, determining a training target weight sequence, determining a weighting loss according to the training target weight sequence and a label corresponding to the historical data of the training candidate object, updating parameters in an object list recommendation model according to the weighting loss in a cyclic iteration mode, thereby improving the feature fusion dimension and the feature vector information density, the method improves the richness of the feature information, increases the information integration of candidate objects, improves the interaction probability of the candidate objects, improves the robustness of the object list recommendation model, displays the objects and the candidate object list through the weight sequence, increases selectable objects of users, improves the practicability and expands the application range of the model.
In some embodiments, after updating the parameters in the object list recommendation model according to the weighted loss by way of loop iteration, further comprising: acquiring current behavior data of a user on an object or a candidate object in a first target object list; and updating the object list recommendation model according to the current behavior data.
Specifically, the current behavior data may be an interaction behavior corresponding to the user login information of the recommendation system, for example, when the user logs in the recommendation system and browses, the interaction behavior may be searching, collecting, purchasing, forwarding, or the like, which is not limited herein.
The object list recommendation model is updated according to the current behavior data, the current behavior data can be used as a parameter for calculating loss, gradient descent is used for minimizing a loss function, the current behavior data is used as a parameter to be transmitted to the loss function in the optimization process, and the current behavior data is carried into the object list recommendation model again according to the gradient update model parameter of the loss function to update the parameter iteratively.
According to the technical scheme provided by the embodiment of the disclosure, the real-time performance of the object list recommendation model is improved and the accuracy of the object list recommendation model is improved by acquiring the current behavior data of the user on the object or the candidate object in the first target object list and updating the object list recommendation model according to the current behavior data.
Fig. 3 is a schematic structural diagram of an object list recommendation model according to an embodiment of the present disclosure, and as shown in fig. 3, the schematic structural diagram of the object list recommendation model includes:
The correlation normalization module 301 is configured to perform a vector normalization process according to at least one candidate object historical feature vector and a weighting coefficient corresponding to the candidate object historical feature vector, and determine a correlation normalization vector corresponding to the candidate object historical data.
And the fusion splicing module 302 is configured to obtain a cross feature vector corresponding to the historical behavior data of the object by fusion splicing the historical behavior feature vector and a related normalized vector corresponding to the historical data of the candidate object.
And the weighting and splicing module 303 is configured to perform weighting and splicing on the cross feature vector corresponding to the historical behavior data of the object and the candidate object attribute feature vector to obtain an object interest feature vector corresponding to the historical behavior data of the object.
The activation weighting module 304 is configured to perform activation weighting according to an object interest feature vector corresponding to the historical behavior data of the object and an environment information feature vector corresponding to the historical environment information data, and determine a target object list.
According to the technical scheme provided by the embodiment of the disclosure, by acquiring the historical data of at least one candidate object associated with the historical behavior data of the object, extracting the features of the historical data to obtain a corresponding historical behavior feature vector and a candidate object historical feature vector, performing weight distribution on the historical behavior feature vector and the candidate object historical feature vector to obtain a weighting coefficient corresponding to each candidate object historical feature vector, normalizing the weighting coefficient and the corresponding candidate historical feature vector by a correlation normalization module 301 to obtain a correlation normalization vector corresponding to the historical data of the candidate object, performing fusion splicing on the correlation normalization vector and the historical behavior feature vector by a fusion splicing module 302 to obtain a cross feature vector, the cross feature vector and the candidate object attribute feature vector are subjected to weighted splicing through the weighted splicing module 303 to obtain an object interest feature vector, the object interest feature vector is activated through an activation function in the activated weighting module 304, and the activation result is subjected to weighted processing to obtain a first target object list, wherein the first target object list is used for displaying the recommended objects and the sequence of all the objects to a user, so that the dimension of feature fusion is improved, the information density of the feature vector is improved, the richness of feature information is improved, the information integration of the candidate objects is improved, the probability of the candidate objects being interacted is improved, the robustness of the object list recommendation model is improved, the selectable objects of the user are increased, the practicability is improved, and the application range is enlarged.
Any combination of the above-mentioned optional solutions may be adopted to form an optional embodiment of the present disclosure, which is not described herein in detail.
The following are device embodiments of the present disclosure that may be used to perform method embodiments of the present disclosure. For details not disclosed in the embodiments of the apparatus of the present disclosure, please refer to the embodiments of the method of the present disclosure.
Fig. 4 is a schematic diagram of an object list recommendation apparatus provided in an embodiment of the present disclosure. As shown in fig. 4, the object list recommending apparatus includes:
An obtaining module 401, configured to obtain, according to the historical behavior data of the object, historical data of at least one candidate object associated with the historical behavior data of the object;
The first processing module 402 is configured to perform feature extraction on the historical behavior data of the object and the historical data of the candidate object, so as to obtain a historical behavior feature vector corresponding to the historical behavior data of the object and a candidate object historical feature vector corresponding to the historical data of the candidate object;
The second processing module 403 is configured to perform weight distribution on the historical behavior feature vector and at least one candidate object historical feature vector, so as to obtain a weighting coefficient corresponding to the candidate object historical feature vector;
A third processing module 404, configured to perform vector normalization processing according to at least one candidate object historical feature vector and a weighting coefficient corresponding to the candidate object historical feature vector, and determine a relevant normalized vector corresponding to the candidate object historical data;
A fourth processing module 405, configured to obtain a cross feature vector corresponding to the historical behavior data of the object by performing fusion and concatenation on the historical behavior feature vector and a related normalized vector corresponding to the historical data of the candidate object;
A fifth processing module 406, configured to perform weighted concatenation on the cross feature vector corresponding to the historical behavior data of the object and the candidate object attribute feature vector, to obtain an object interest feature vector corresponding to the historical behavior data of the object;
A first determining module 407, configured to perform activation weighting according to an object interest feature vector corresponding to historical behavior data of an object, and determine a first target weight sequence, where the first target weight sequence includes a weight value corresponding to the historical behavior data of the object and a weight value corresponding to the historical data of the candidate object;
the second determining module 408 is configured to determine, according to the first target weight sequence, a first target object list recommended to the user, where the first target object list includes an object and at least one candidate object.
According to the technical scheme provided by the embodiment of the disclosure, by acquiring the historical data of at least one candidate object associated with the historical behavior data of the object, performing feature extraction on the data to obtain a corresponding historical behavior feature vector and a candidate object historical feature vector, performing weight distribution on the historical behavior feature vector and the candidate object historical feature vector to obtain a weighting coefficient corresponding to each candidate object historical feature vector, normalizing the weighting coefficient and the corresponding candidate historical feature vector to obtain a related normalized vector corresponding to the historical data of the candidate object, performing fusion splicing on the related normalized vector and the historical behavior feature vector to obtain a cross feature vector, performing weighted splicing on the cross feature vector and the candidate object attribute feature vector to obtain an object interest feature vector, activating the object interest feature vector through an activating function, performing weighted processing on an activating result to obtain a first target weight sequence, and determining a first target weight sequence according to the first target weight sequence.
In some embodiments, the object list recommending apparatus is further configured to obtain historical environmental information data, where the historical environmental information data is used to indicate historical environmental information of terminal devices corresponding to the object and the candidate object; extracting features of the historical environmental information data to obtain environmental information feature vectors corresponding to the historical environmental information data; activating and weighting according to object interest feature vectors corresponding to the historical behavior data of the object and environment information feature vectors corresponding to the historical environment information data, and determining a second target weight sequence, wherein the second target weight sequence comprises weight values corresponding to the historical behavior data of the object and weight values corresponding to the historical data of the candidate object; and determining a second target object list recommended to the user according to the second target weight sequence, wherein the ordering of the objects and the candidate objects in the second target object list and the ordering of the objects and the candidate objects in the first target object list are the same or different.
In some embodiments, the third processing module 404 is specifically configured to perform exponential normalization processing on the weighting coefficient corresponding to each candidate object history feature vector according to at least one candidate object history feature vector, to obtain a normalization result corresponding to each candidate object history feature vector; adding the normalization result corresponding to each candidate object history feature vector to obtain a normalization addition result; dividing the normalization result and the normalization summation result corresponding to each candidate object history feature vector to obtain a weighting coefficient corresponding to each candidate object history feature vector; and determining a relevant normalized vector corresponding to the historical data of the candidate object according to the weighting coefficient corresponding to the historical feature vector of the candidate object and the historical feature vector of the candidate object.
In some embodiments, the fourth processing module 405 is specifically configured to perform subtraction processing on the historical behavior feature vector and a relevant normalized vector corresponding to the historical data of the candidate object, to obtain a difference result; multiplying the historical behavior feature vector and a related normalized vector corresponding to the historical data of the candidate object to obtain a product result; and according to the preset dimension, carrying out series fusion on the historical behavior feature vector, the related normalized vector corresponding to the historical data of the candidate object, the difference result and the product result to obtain the cross feature vector corresponding to the historical behavior data of the object.
In some embodiments, the fifth processing module 406 is specifically configured to multiply the cross feature vector corresponding to the historical behavior data of the object with the candidate object attribute feature vector, and perform exponential normalization on the processing result to obtain an attribute normalization result; performing self-attention weight processing on the attribute normalization result to obtain an attribute self-attention weight result; and carrying out weighted summation on the attribute self-attention weight result to obtain the object interest feature vector corresponding to the historical behavior data of the object.
In some embodiments, the object list recommending apparatus is further configured to obtain historical behavior data of a training object, historical data of a training candidate object, a tag corresponding to the historical data of the training candidate object, attribute data of the training candidate object, and historical training environment information data, where the tag corresponding to the historical data of the training candidate object is used to indicate a real weight value corresponding to the historical data of the training candidate object; inputting the historical behavior data of the training object, the historical data of the training candidate object, the attribute data of the training candidate object and the historical environmental information data of the training history to the object list recommendation model for feature extraction to obtain a historical behavior training vector corresponding to the historical behavior data of the training object, a candidate object historical training vector corresponding to the historical data of the training candidate object, a candidate object attribute training vector corresponding to the attribute data of the training candidate object and a historical environmental information training vector corresponding to the historical environmental information data of the training candidate object; performing weight distribution on the historical behavior training vector and at least one candidate object historical training vector to obtain a weighting coefficient corresponding to the candidate object historical training vector; carrying out vector normalization processing according to at least one candidate object historical training vector and a weighting coefficient corresponding to the candidate object historical training vector, and determining a relevant normalization vector corresponding to the training candidate object data; the cross feature vectors corresponding to the historical behavior data of the training object are obtained by fusion and splicing of the historical behavior training vectors and the related normalization vectors corresponding to the data of the training candidate object; performing weighted concatenation on the cross feature vector corresponding to the historical behavior data of the training object and the candidate object attribute training vector to obtain an object interest training vector corresponding to the historical behavior data of the training object; activating and weighting according to an object interest training vector corresponding to the historical behavior data of the training object and an environment information training vector, and determining a training target weight sequence, wherein the training target weight sequence comprises a weight value corresponding to the historical behavior data of the training object and a weight value corresponding to the historical data of the training candidate object, and determining a weighting loss according to the training target weight sequence and a label corresponding to the historical data of the training candidate object; and updating parameters in the object list recommendation model according to the weighting loss in a cyclic iteration mode.
In some embodiments, updating parameters in the object list recommendation model according to the weighted loss is further used to obtain current behavior data of the user on the object or candidate object in the first target object list; and updating the object list recommendation model according to the current behavior data.
It should be understood that the sequence number of each step in the foregoing embodiment does not mean that the execution sequence of each process should be determined by the function and the internal logic of each process, and should not constitute any limitation on the implementation process of the embodiments of the disclosure.
Fig. 5 is a schematic diagram of an electronic device 5 provided by an embodiment of the present disclosure. As shown in fig. 5, the electronic apparatus 5 of this embodiment includes: a processor 501, a memory 502 and a computer program 503 stored in the memory 502 and executable on the processor 501. The steps of the various method embodiments described above are implemented by processor 501 when executing computer program 503. Or the processor 501 when executing the computer program 503 performs the functions of the modules/units in the above-described apparatus embodiments.
The electronic device 5 may be a desktop computer, a notebook computer, a palm computer, a cloud server, or the like. The electronic device 5 may include, but is not limited to, a processor 501 and a memory 502. It will be appreciated by those skilled in the art that fig. 5 is merely an example of the electronic device 5 and is not limiting of the electronic device 5 and may include more or fewer components than shown, or different components.
The processor 501 may be a central processing unit (Central Processing Unit, CPU), but may also be other general purpose processors, digital signal processors (DIGITAL SIGNAL processor, DSP), application SPECIFIC INTEGRATED Circuit (ASIC), field-programmable gate array (field-programmable GATE ARRAY, FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, etc.
The memory 502 may be an internal storage unit of the electronic device 5, for example, a hard disk or a memory of the electronic device 5. The memory 502 may also be an external storage device of the electronic device 5, such as a plug-in hard disk, a smart memory card (SMART MEDIA CARD, SMC), a Secure Digital (SD) card, a flash memory card (FLASH CARD) or the like, which are provided on the electronic device 5. Memory 502 may also include both internal storage units and external storage devices of electronic device 5. The memory 502 is used to store computer programs and other programs and data required by the electronic device.
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions. The functional units and modules in the embodiment may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit, where the integrated units may be implemented in a form of hardware or a form of a software functional unit.
The integrated modules/units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a readable storage medium (e.g., a computer readable storage medium). Based on such understanding, the present disclosure may implement all or part of the flow of the method of the above-described embodiments, or may be implemented by a computer program to instruct related hardware, and the computer program may be stored in a computer readable storage medium, where the computer program, when executed by a processor, may implement the steps of the method embodiments described above. The computer program may comprise computer program code, which may be in source code form, object code form, executable file or in some intermediate form, etc. The computer readable storage medium may include: any entity or device capable of carrying computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (Random Access Memory, RAM), an electrical carrier signal, a telecommunications signal, a software distribution medium, and so forth.
The above embodiments are merely for illustrating the technical solution of the present disclosure, and are not limiting thereof; although the present disclosure has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the disclosure, and are intended to be included in the scope of the present disclosure.
Claims (10)
1. An object list recommendation method, comprising:
according to the historical behavior data of the object, acquiring the historical data of at least one candidate object associated with the historical behavior data of the object;
Extracting features of the historical behavior data of the object and the historical data of the candidate object to obtain a historical behavior feature vector corresponding to the historical behavior data of the object and a candidate object historical feature vector corresponding to the historical data of the candidate object;
Performing weight distribution on the historical behavior feature vector and at least one candidate object historical feature vector to obtain a weighting coefficient corresponding to the candidate object historical feature vector;
Vector normalization processing is carried out according to at least one candidate object history feature vector and a weighting coefficient corresponding to the candidate object history feature vector, and a relevant normalization vector corresponding to the candidate object history data is determined;
the cross feature vector corresponding to the historical behavior data of the object is obtained by fusion and splicing of the historical behavior feature vector and the related normalized vector corresponding to the historical data of the candidate object;
Performing weighted splicing on the cross feature vector corresponding to the historical behavior data of the object and the candidate object attribute feature vector to obtain an object interest feature vector corresponding to the historical behavior data of the object;
Performing activation weighting according to object interest feature vectors corresponding to the historical behavior data of the object, and determining a first target weight sequence, wherein the first target weight sequence comprises a weight value corresponding to the historical behavior data of the object and a weight value corresponding to the historical data of the candidate object;
And determining a first target object list recommended to the user according to the first target weight sequence, wherein the first target object list comprises the object and at least one candidate object.
2. The object list recommendation method according to claim 1, further comprising, before the activation weighting is performed on the object interest feature vector corresponding to the historical behavior data of the object, determining a first target weight sequence:
Acquiring historical environment information data, wherein the historical environment information data is used for indicating historical environment information of the object and terminal equipment corresponding to the candidate object;
Extracting features of the historical environmental information data to obtain environmental information feature vectors corresponding to the historical environmental information data;
Performing activation weighting according to an object interest feature vector corresponding to the historical behavior data of the object and an environment information feature vector corresponding to the historical environment information data, and determining a second target weight sequence, wherein the second target weight sequence comprises a weight value corresponding to the historical behavior data of the object and a weight value corresponding to the historical data of the candidate object;
and determining a second target object list recommended to the user according to the second target weight sequence, wherein the ordering of the objects and the candidate objects in the second target object list and the ordering of the objects and the candidate objects in the first target object list are the same or different.
3. The method according to claim 1, wherein the performing vector normalization processing according to at least one of the candidate object history feature vector and a weighting coefficient corresponding to the candidate object history feature vector, and determining a relevant normalized vector corresponding to the candidate object history feature vector, includes:
according to at least one candidate object history feature vector, carrying out exponential normalization processing on a weighting coefficient corresponding to each candidate object history feature vector to obtain a normalization result corresponding to each candidate object history feature vector;
Adding the normalization result corresponding to each candidate object history feature vector to obtain a normalization addition result;
Dividing the normalization result corresponding to each candidate object history feature vector and the normalization summation result to obtain a weighting coefficient corresponding to each candidate object history feature vector;
And determining a relevant normalized vector corresponding to the historical data of the candidate object according to the weighting coefficient corresponding to the historical feature vector of the candidate object and the historical feature vector of the candidate object.
4. The method for recommending an object list according to claim 1, wherein the obtaining the cross feature vector corresponding to the historical behavior data of the object by fusion splicing the historical behavior feature vector and the related normalized vector corresponding to the historical data of the candidate object includes:
Subtracting the historical behavior feature vector from a relevant normalized vector corresponding to the historical data of the candidate object to obtain a difference result;
Multiplying the historical behavior feature vector and a related normalized vector corresponding to the historical data of the candidate object to obtain a product result;
And according to a preset dimension, carrying out series fusion on the historical behavior feature vector, the related normalized vector corresponding to the historical data of the candidate object, the difference result and the product result to obtain the cross feature vector corresponding to the historical behavior data of the object.
5. The method for recommending an object list according to claim 1, wherein the weighting and stitching the cross feature vector corresponding to the historical behavior data of the object and the candidate object attribute feature vector to obtain the object interest feature vector corresponding to the historical behavior data of the object, includes:
Multiplying the cross feature vector corresponding to the historical behavior data of the object with the candidate object attribute feature vector, and carrying out index normalization on the processing result to obtain an attribute normalization result;
performing self-attention weight processing on the attribute normalization result to obtain an attribute self-attention weight result;
And carrying out weighted summation on the attribute self-attention weight result to obtain an object interest feature vector corresponding to the historical behavior data of the object.
6. The object list recommendation method according to claim 1, further comprising, before the acquiring, from the historical behavior data of the object, the historical data of at least one candidate object associated with the historical behavior data of the object:
Acquiring historical behavior data of a training object, historical data of a training candidate object, labels corresponding to the historical data of the training candidate object, attribute data of the training candidate object and training historical environment information data, wherein the labels corresponding to the historical data of the training candidate object are used for indicating real weight values corresponding to the historical data of the training candidate object;
Inputting the historical behavior data of the training object, the historical data of the training candidate object, the attribute data of the training candidate object and the historical environment information data of the training to an object list recommendation model for feature extraction to obtain a historical behavior training vector corresponding to the historical behavior data of the training object, a candidate object historical training vector corresponding to the historical data of the training candidate object, a candidate object attribute training vector corresponding to the attribute data of the training candidate object and a historical environment information training vector corresponding to the historical environment information data;
Performing weight distribution on the historical behavior training vector and at least one candidate object historical training vector to obtain a weighting coefficient corresponding to the candidate object historical training vector;
Vector normalization processing is carried out according to at least one candidate object historical training vector and a weighting coefficient corresponding to the candidate object historical training vector, and a relevant normalization vector corresponding to the training candidate object data is determined;
The cross feature vector corresponding to the historical behavior data of the training object is obtained by fusion and splicing of the historical behavior training vector and the related normalization vector corresponding to the data of the training candidate object;
Performing weighted concatenation on the cross feature vector corresponding to the historical behavior data of the training object and the candidate object attribute training vector to obtain an object interest training vector corresponding to the historical behavior data of the training object;
Performing activation weighting according to an object interest training vector corresponding to the historical behavior data of the training object and an environment information training vector, determining a training target weight sequence, wherein the training target weight sequence comprises a weight value corresponding to the historical behavior data of the training object and a weight value corresponding to the historical data of the training candidate object, and determining a weighting loss according to the training target weight sequence and a label corresponding to the historical data of the training candidate object;
and updating parameters in the object list recommendation model according to the weighted loss in a cyclic iteration mode.
7. The object list recommendation method according to claim 6, further comprising, after updating parameters in an object list recommendation model according to the weight loss in the loop iteration manner:
Acquiring current behavior data of the user on the object or the candidate object in the first target object list;
and updating the object list recommendation model according to the current behavior data.
8. An object list recommendation apparatus, comprising:
the acquisition module is used for acquiring the historical data of at least one candidate object associated with the historical behavior data of the object according to the historical behavior data of the object;
the first processing module is used for extracting characteristics of the historical behavior data of the object and the historical data of the candidate object to obtain a historical behavior characteristic vector corresponding to the historical behavior data of the object and a candidate object historical characteristic vector corresponding to the historical data of the candidate object;
The second processing module is used for carrying out weight distribution on the historical behavior feature vector and at least one candidate object historical feature vector to obtain a weighting coefficient corresponding to the candidate object historical feature vector;
The third processing module is used for carrying out vector normalization processing according to at least one candidate object history feature vector and a weighting coefficient corresponding to the candidate object history feature vector, and determining a relevant normalization vector corresponding to the candidate object history data;
the fourth processing module is used for obtaining a cross feature vector corresponding to the historical behavior data of the object by fusing and splicing the historical behavior feature vector and a related normalized vector corresponding to the historical data of the candidate object;
A fifth processing module, configured to perform weighted concatenation on a cross feature vector corresponding to the historical behavior data of the object and a candidate object attribute feature vector, so as to obtain an object interest feature vector corresponding to the historical behavior data of the object;
The first determining module is used for performing activation weighting according to the object interest feature vector corresponding to the historical behavior data of the object, and determining a first target weight sequence, wherein the first target weight sequence comprises a weight value corresponding to the historical behavior data of the object and a weight value corresponding to the historical data of the candidate object;
And the second determining module is used for determining a first target object list recommended to the user according to the first target weight sequence, wherein the first target object list comprises the object and at least one candidate object.
9. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1 to 7 when the computer program is executed.
10. A readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the steps of the method according to any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311864492.4A CN117893280A (en) | 2023-12-29 | 2023-12-29 | Object list recommendation method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311864492.4A CN117893280A (en) | 2023-12-29 | 2023-12-29 | Object list recommendation method and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117893280A true CN117893280A (en) | 2024-04-16 |
Family
ID=90642071
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311864492.4A Pending CN117893280A (en) | 2023-12-29 | 2023-12-29 | Object list recommendation method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117893280A (en) |
-
2023
- 2023-12-29 CN CN202311864492.4A patent/CN117893280A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109492772B (en) | Method and device for generating information | |
CN111125574B (en) | Method and device for generating information | |
CN113781149A (en) | Information recommendation method and device, computer-readable storage medium and electronic equipment | |
WO2022156589A1 (en) | Method and device for determining live broadcast click rate | |
CN113822734B (en) | Method and device for generating information | |
CN112749323A (en) | Method and device for constructing user portrait | |
CN113792952A (en) | Method and apparatus for generating a model | |
CN113450172B (en) | Commodity recommendation method and device | |
CN111768218B (en) | Method and device for processing user interaction information | |
CN116186541A (en) | Training method and device for recommendation model | |
CN114429384B (en) | Intelligent product recommendation method and system based on e-commerce platform | |
CN111125502A (en) | Method and apparatus for generating information | |
CN112328899B (en) | Information processing method, information processing apparatus, storage medium, and electronic device | |
CN113327145B (en) | Article recommendation method and device | |
CN113313542B (en) | Method and device for pushing channel pages | |
CN117893280A (en) | Object list recommendation method and device | |
CN113837843A (en) | Product recommendation method, device, medium and electronic equipment | |
CN110738538B (en) | Method and device for identifying similar objects | |
CN113449175A (en) | Hot data recommendation method and device | |
Sudarsan et al. | E-commerce Website with Image Search and Price Prediction | |
CN113159877A (en) | Data processing method, device, system and computer readable storage medium | |
CN116911912B (en) | Method and device for predicting interaction objects and interaction results | |
Diqi | Deeprec: Efficient product recommendation model for e-commerce using cnn | |
CN113793160B (en) | Put-in data processing method, device, equipment and storage medium | |
CN118013113A (en) | Object list recommendation method and device, electronic equipment and readable storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |