CN116308634A - Double-tower model recommendation method and device based on behavior sequence and weight sharing - Google Patents
Double-tower model recommendation method and device based on behavior sequence and weight sharing Download PDFInfo
- Publication number
- CN116308634A CN116308634A CN202310136544.XA CN202310136544A CN116308634A CN 116308634 A CN116308634 A CN 116308634A CN 202310136544 A CN202310136544 A CN 202310136544A CN 116308634 A CN116308634 A CN 116308634A
- Authority
- CN
- China
- Prior art keywords
- user
- vector
- article
- target user
- recommended
- 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 58
- 239000013598 vector Substances 0.000 claims abstract description 302
- 238000012545 processing Methods 0.000 claims abstract description 46
- 230000006399 behavior Effects 0.000 claims description 103
- 238000004590 computer program Methods 0.000 claims description 16
- 230000009977 dual effect Effects 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 11
- 238000007906 compression Methods 0.000 claims description 10
- 230000006835 compression Effects 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 8
- 238000013528 artificial neural network Methods 0.000 claims description 7
- 230000003542 behavioural effect Effects 0.000 claims description 5
- 239000010410 layer Substances 0.000 description 31
- 238000001994 activation Methods 0.000 description 9
- 230000008569 process Effects 0.000 description 9
- 238000004891 communication Methods 0.000 description 8
- 238000004422 calculation algorithm Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 239000002356 single layer Substances 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 3
- 230000001960 triggered effect Effects 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 238000003491 array Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000001953 sensory effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
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)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- General Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a double-tower model recommendation method and device based on behavior sequence and weight sharing, comprising the following steps: acquiring at least one user characteristic corresponding to a target user, and processing each user characteristic through a SENET network to obtain a user characteristic vector matched with the target user; according to the behavior sequence corresponding to the target user, at least one associated article corresponding to the target user is obtained, and a behavior feature vector matched with the target user is determined according to each associated article; acquiring article characteristics corresponding to a plurality of articles to be recommended respectively, and processing the article characteristics through a SENet network to obtain article vectors matched with the articles to be recommended; and determining a target item in the plurality of items to be recommended according to the user feature vector, the behavior feature vector and the item vector, and recommending the target item to the target user. The technical scheme of the embodiment of the invention can improve the accuracy of the article recommendation result.
Description
Technical Field
The invention relates to the technical field of computers, in particular to a double-tower model recommendation method and device based on behavior sequence and weight sharing.
Background
With the rapid development of internet technology, more and more users select and purchase items through a network. Because of the wide variety and number of items in application platforms, it is becoming particularly important how to provide personalized item recommendations to users.
In the prior art, an Embedding algorithm is used as a feature coding mode, and is widely applied to the current recommendation system. Typically, an Embedding algorithm can be used to recall vector models, such as a deep structured semantic model (Deep Structured Semantic Models, DSSM). The DSSM model can input the user characteristics and the object characteristics into the respective neural networks respectively, acquire the ebedding vector representation of the user characteristics and the object characteristics, and then calculate the correlation degree between the user and the object through the vector cosine similarity.
However, because the DSSM model needs to build independent network structures corresponding to the user and the article respectively, the model is difficult to effectively capture the cross features between the user and the article, and therefore the accuracy of the article recommendation result is low; secondly, because the cross feature can generate larger calculation amount, even if the DSSM model acquires the cross feature, the offline recall performance of the DSSM model can be influenced.
Disclosure of Invention
The invention provides a double-tower model recommendation method and a double-tower model recommendation device based on behavior sequence and weight sharing, which can improve the accuracy of article recommendation results.
According to an aspect of the invention, there is provided a dual-tower model recommendation method based on behavior sequence and weight sharing, the method comprising:
acquiring at least one user characteristic corresponding to a target user, and processing each user characteristic through a compression activation network SENet to obtain a user characteristic vector matched with the target user;
according to the behavior sequence corresponding to the target user, at least one associated article corresponding to the target user is obtained, and according to each associated article, a behavior feature vector matched with the target user is determined;
acquiring article characteristics corresponding to a plurality of articles to be recommended respectively, and processing the article characteristics through a SENet network to obtain article vectors matched with the articles to be recommended;
and determining a target object in the plurality of objects to be recommended according to the user feature vector, the behavior feature vector and the object vector, and recommending the target object to a target user.
Optionally, processing each user feature through the compression activation network SENet to obtain a user feature vector matched with the target user, including:
Coding each user characteristic to obtain a sparse characteristic vector corresponding to a target user;
inputting the sparse feature vector into an embedded layer, and processing the sparse feature vector through the embedded layer to obtain a dense feature vector corresponding to a target user;
and inputting the dense feature vector into a SENET network, and processing the dense feature vector through the SENET network to obtain a user feature vector matched with the target user.
Optionally, determining, according to each associated article, a behavior feature vector matched with the target user, including:
acquiring article characteristics corresponding to each associated article respectively, and generating sparse characteristic vectors corresponding to each associated article respectively according to the article characteristics;
the sparse feature vectors corresponding to the associated articles are input into an embedding layer, and the sparse feature vectors are processed through the embedding layer to obtain dense feature vectors corresponding to the associated articles;
and carrying out mean value pooling treatment on the dense feature vectors corresponding to all the associated objects to obtain a behavior feature vector matched with the target user.
Optionally, processing each item feature through a SENet network to obtain an item vector matched with each item to be recommended, including:
Inputting the dense feature vectors corresponding to the articles to be recommended into a SENet network, and processing the dense feature vectors through the SENet network to obtain article feature vectors matched with the articles to be recommended;
and inputting the item feature vectors into a dynamic neural network DNN to obtain item vectors matched with the items to be recommended.
Optionally, determining the target item from the plurality of items to be recommended according to the user feature vector, the behavior feature vector and the item vector includes:
determining a user vector corresponding to the target user according to the user feature vector and the behavior feature vector;
according to the user vector and each item vector, determining the correlation degree between the target user and each item to be recommended;
and determining the target object in the plurality of objects to be recommended according to the correlation.
Optionally, determining the user vector corresponding to the target user according to the user feature vector and the behavior feature vector includes:
and splicing the user characteristic vector and the behavior characteristic vector, and inputting a splicing result into a DNN network to obtain a user vector corresponding to the target user.
Optionally, determining the correlation between the target user and each item to be recommended according to the user vector and each item vector includes:
And calculating cosine similarity between the user vector and each article vector, and determining the correlation between the target user and each article to be recommended according to the cosine similarity.
According to another aspect of the present invention, there is provided a dual tower model recommendation apparatus based on behavior sequence and weight sharing, the apparatus comprising:
the user characteristic determining module is used for acquiring at least one user characteristic corresponding to the target user, and processing each user characteristic through the compression activation network SENet to obtain a user characteristic vector matched with the target user;
the behavior feature determining module is used for acquiring at least one associated article corresponding to the target user according to the behavior sequence corresponding to the target user, and determining a behavior feature vector matched with the target user according to each associated article;
the article vector determining module is used for acquiring article characteristics corresponding to a plurality of articles to be recommended respectively, and processing the article characteristics through a SENet network to obtain article vectors matched with the articles to be recommended;
and the target article determining module is used for determining a target article in a plurality of articles to be recommended according to the user characteristic vector, the behavior characteristic vector and the article vector and recommending the target article to a target user.
According to another aspect of the present invention, there is provided an electronic apparatus including:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein,,
the memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the dual tower model recommendation method based on behavior sequence and weight sharing according to any of the embodiments of the present invention.
According to another aspect of the present invention, there is provided a computer readable storage medium storing computer instructions for causing a processor to implement the dual tower model recommendation method based on behavior sequence and weight sharing according to any of the embodiments of the present invention when executed.
According to the technical scheme provided by the embodiment of the invention, the user characteristics corresponding to the target user are obtained through obtaining at least one user characteristic, the user characteristics are processed through the SENET network to obtain the user characteristic vector matched with the target user, at least one associated article corresponding to the target user is obtained according to the behavior sequence corresponding to the target user, the behavior characteristic vector matched with the target user is determined according to the associated articles, the article characteristics respectively corresponding to a plurality of articles to be recommended are obtained, the article characteristics are processed through the SENET network to obtain the article vector matched with the articles to be recommended, the target article is determined in the articles to be recommended according to the user characteristic vector, the behavior characteristic vector and the article vector, and the target article is recommended to the target user.
It should be understood that the description in this section is not intended to identify key or critical features of the embodiments of the invention or to delineate the scope of the invention. Other features of the present invention will become apparent from the description that follows.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a dual tower model recommendation method based on behavior sequence and weight sharing provided in accordance with an embodiment of the present invention;
FIG. 2 is a flow chart of another dual tower model recommendation method based on behavior sequence and weight sharing provided in accordance with an embodiment of the present invention;
FIG. 3a is a flow chart of another dual tower model recommendation method based on behavior sequence and weight sharing provided in accordance with an embodiment of the present invention;
FIG. 3b is a flow chart of another dual tower model recommendation method based on behavior sequence and weight sharing provided in accordance with an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of a dual-tower model recommendation device based on behavior sequence and weight sharing according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an electronic device implementing a dual-tower model recommendation method based on behavior sequence and weight sharing according to an embodiment of the present invention.
Detailed Description
In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the invention described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
Fig. 1 is a flowchart of a dual-tower model recommendation method based on behavior sequence and weight sharing, which is provided in an embodiment of the present invention, and the embodiment may be suitable for a case that an application platform recommends related items to a user, where the method may be performed by a dual-tower model recommendation device based on behavior sequence and weight sharing, and the device may be implemented in a form of hardware and/or software, and the device may be configured in an electronic device (e.g., a terminal or a server) with a data processing function. As shown in fig. 1, the method includes:
In this embodiment, the target user may be a login user of the object to be recommended in the application platform. Optionally, after detecting that the target user logs in the application platform, the statistical features of the sex, the age, the region, the preference category, the preference brand, the liveness, the value and the like of the target user can be obtained, and the statistical features are used as the user features of the target user.
In this step, after the user feature corresponding to the target user is obtained, the user feature may be processed through a compressed activation network (Squeeze-and-Excitation Networks, SENet) to obtain a user feature vector.
In a specific embodiment, the SENet network can be divided into two processes, compression (squeeze) and activation (specification). The compression process is to average the user characteristics and compress each user characteristic into a one-dimensional value z i For specific processes, reference may be made to the following formula:
wherein k can be expressed as a feature quantity, v i May be represented as specific user characteristics.
The activation process refers to cross processing of input user features through a multi-layer sensor (Multilayer Perceptron, MLP) and outputting corresponding weight values, and the specific processing process can refer to the following formula:
S＝F ex (Z，W)＝θ(W 2 θ(W 1 Z))
and 120, acquiring at least one associated article corresponding to the target user according to the behavior sequence corresponding to the target user, and determining a behavior feature vector matched with the target user according to each associated article.
In this embodiment, the behavior sequence may include a plurality of behaviors triggered by the target user in the application platform, such as browsing, clicking, adding shopping carts, collecting, purchasing, and the like. Optionally, before generating the behavior sequence, the behaviors triggered by the target user may be screened, and the screening result may be added to the behavior sequence.
In a specific embodiment, all click behaviors triggered by the target user in the application platform can be obtained, and according to the page stay time corresponding to the click behaviors, the click behaviors meeting the requirements are screened and added into the behavior sequence. The minimum threshold value corresponding to the page residence time may be set to 15s, and the specific value may be adjusted according to the actual situation, which is not limited in this embodiment.
In this step, the item closest to the current time, which is involved in the behavior sequence, may be regarded as the associated item corresponding to the target user. After the associated item is obtained, item features of the associated item, such as item category, item brand, item ID, item price, etc., may be extracted, and then a behavior feature vector may be generated based on the item features of all of the associated items.
And 130, acquiring article characteristics corresponding to a plurality of articles to be recommended respectively, and processing the article characteristics through a SENet network to obtain article vectors matched with the articles to be recommended.
In this embodiment, optionally, the item features corresponding to the item to be recommended may include an item category, an item brand, an item ID, an item price, and the like.
And 140, determining a target item in a plurality of items to be recommended according to the user characteristic vector, the behavior characteristic vector and the item vector, and recommending the target item to a target user.
In this embodiment, optionally, the relevance between the target user and each item to be recommended may be determined according to the user feature vector, the behavior feature vector, and the item to be recommended with higher relevance may be used as the target item, and then the target item is recommended to the target user.
In this embodiment, by adding the SENet network structure to the recommendation system, feedback of important features can be enhanced, interference of non-important features can be suppressed, and validity of cross features between users and articles can be ensured by respectively suppressing secondary features in the user layer and the article layer, so that accuracy of article recommendation results can be improved.
According to the technical scheme provided by the embodiment of the invention, the user characteristics corresponding to the target user are obtained through obtaining at least one user characteristic, the user characteristics are processed through the SENET network to obtain the user characteristic vector matched with the target user, at least one associated article corresponding to the target user is obtained according to the behavior sequence corresponding to the target user, the behavior characteristic vector matched with the target user is determined according to the associated articles, the article characteristics respectively corresponding to a plurality of articles to be recommended are obtained, the article characteristics are processed through the SENET network to obtain the article vector matched with the articles to be recommended, the target article is determined in the articles to be recommended according to the user characteristic vector, the behavior characteristic vector and the article vector, and the target article is recommended to the target user.
Fig. 2 is a flowchart of a dual-tower model recommendation method based on behavior sequence and weight sharing according to a second embodiment of the present invention, which is a further refinement of the foregoing embodiment. As shown in fig. 2, the method includes:
In this embodiment, optionally, the user feature may be encoded by a one-hot text feature extraction algorithm to obtain a sparse feature vector.
In this step, the sparse feature vector may be input into an embedding layer embedding-layer, and the sparse feature vector may be mapped to a dense space by an embedding algorithm to obtain a dense feature vector.
And 203, inputting the dense feature vector into a SENET network, and processing the dense feature vector through the SENET network to obtain a user feature vector matched with the target user.
In this step, dense feature vectors may be input to the SENet network, resulting in weighted user feature vectors.
In this step, the item features corresponding to the associated items may be processed in the same manner as in step 201, to obtain sparse feature vectors corresponding to the associated items.
In this step, the sparse feature vectors may be processed in the same manner as in step 202 to obtain dense feature vectors corresponding to each associated item.
And 207, carrying out mean value pooling processing on the dense feature vectors corresponding to all the associated objects to obtain a behavior feature vector matched with the target user.
In this step, dense feature vectors corresponding to all the associated objects may be input to a Mean pooling layer (Mean pooling) to obtain a behavior feature vector.
In the step, after the article characteristics corresponding to the plurality of articles to be recommended are obtained, the article characteristics can be encoded to obtain sparse characteristic vectors corresponding to the articles to be recommended respectively, then the sparse characteristic vectors are input into an embedding layer to obtain dense characteristic vectors corresponding to the articles to be recommended respectively, and finally the dense characteristic vectors are input into a SENET network to be weighted dynamically to obtain article characteristic vectors matched with the articles to be recommended.
In this step, each item feature vector may be input into a single-layer dynamic neural network (Dynamic Neural Network, DNN), and the item feature vector is processed by the single-layer DNN to obtain an item vector matching each item to be recommended.
And step 210, determining a target item in a plurality of items to be recommended according to the user characteristic vector, the behavior characteristic vector and the item vector, and recommending the target item to a target user.
In a specific embodiment, the item recommendation model determined by the above method may use cross entropy as a loss function, as shown in the following formula:
according to the technical scheme provided by the embodiment of the invention, the sparse feature vectors are obtained by obtaining the user features corresponding to the target user, the sparse feature vectors are input into the embedded layer to obtain the dense feature vectors, the dense feature vectors are input into the SENet network to obtain the user feature vectors, the associated articles and the article features corresponding to the associated articles respectively are obtained according to the behavior sequence of the target user, the sparse feature vectors corresponding to the associated articles respectively are generated according to the article features, the sparse feature vectors corresponding to the associated articles are input into the embedded layer to obtain the dense feature vectors corresponding to the associated articles respectively, the dense feature vectors corresponding to all the associated articles are subjected to mean pooling treatment to obtain the behavior feature vectors, the article features corresponding to the articles to be recommended are obtained, the dense feature vectors corresponding to the articles to be recommended are input into the SENet network to obtain the article feature vectors matched with the articles to be recommended, the article feature vectors are input into the DNN network to obtain the article vectors matched with the articles to be recommended, the target articles are determined according to the user feature vectors, the behavior feature vectors and the articles to be recommended to the articles, and the technical means that the target articles are recommended to the target user can improve the accuracy of the article recommendation.
Fig. 3a is a flowchart of a dual-tower model recommendation method based on behavior sequence and weight sharing according to a third embodiment of the present invention, which is a further refinement of the foregoing embodiment. As shown in fig. 3a, the method comprises:
And 330, acquiring article characteristics corresponding to the plurality of articles to be recommended respectively, and processing the article characteristics through a SENet network to obtain article vectors matched with the articles to be recommended.
In this embodiment, optionally, a preset linear or nonlinear algorithm may be used to process the user feature vector and the behavior feature vector to obtain the user vector.
In one implementation manner of this embodiment, determining, according to the user feature vector and the behavior feature vector, a user vector corresponding to the target user includes: and splicing the user characteristic vector and the behavior characteristic vector, and inputting a splicing result into a DNN network to obtain a user vector corresponding to the target user.
In a specific embodiment, the user feature vector and the behavior feature vector may be input together into a splice layer Concat, the user feature vector and the behavior feature vector are spliced by the Concat layer, and then the splicing result is input into a single-layer DNN network to obtain the user vector.
And 350, determining the correlation degree between the target user and each item to be recommended according to the user vector and each item vector.
In one implementation manner of the present embodiment, determining the relevance between the target user and each item to be recommended according to the user vector and each item vector includes: and calculating cosine similarity between the user vector and each article vector, and determining the correlation between the target user and each article to be recommended according to the cosine similarity.
In a specific embodiment, the cosine similarity may be directly used as the correlation between the target user and the object to be recommended, or the cosine similarity may be linearly processed, and the processing result may be used as the correlation between the target user and the object to be recommended.
And step 360, determining a target item in a plurality of items to be recommended according to the relevance, and recommending the target item to a target user.
In this step, optionally, the plurality of articles to be recommended may be ranked according to the degree of correlation from large to small, and the articles to be recommended ranked in front may be used as target articles; or an item to be recommended whose correlation degree is greater than a preset threshold value may be regarded as the target item.
According to the technical scheme provided by the embodiment of the invention, the user characteristics corresponding to the target user are obtained through obtaining the user characteristics, the user characteristics are processed through the SENET network to obtain the user characteristic vectors, the associated articles are obtained according to the behavior sequences corresponding to the target user, the behavior characteristic vectors are determined according to the associated articles to obtain the article characteristics corresponding to the articles to be recommended respectively, the article characteristics are processed through the SENET network to obtain the article vectors matched with the articles to be recommended, the user vectors are determined according to the user characteristic vectors and the behavior characteristic vectors, the correlation between the target user and the articles to be recommended is determined according to the user characteristic vectors and the article vectors, and the technical means that the target articles are recommended to the target user are determined according to the correlation, so that the accuracy of article recommendation results can be improved.
On the basis of the above embodiment, the present embodiment further provides a preferred implementation manner of a dual-tower model recommendation method based on behavior sequence and weight sharing, as shown in fig. 3b, where the method includes the following steps:
step 1: acquiring User characteristics (User characteristics) corresponding to a target User, coding the User characteristics to obtain sparse characteristic vectors, inputting the sparse characteristic vectors into an Embedding layer (Embedding), processing the sparse characteristic vectors through the Embedding layer to obtain dense characteristic vectors, inputting the dense characteristic vectors into a SENET network, and processing the dense characteristic vectors through the SENET network to obtain User characteristic vectors matched with the target User;
step 2: according to a behavior sequence corresponding to a target user, acquiring article characteristics (item 1, item2 and item3 … …) corresponding to the associated articles respectively, generating sparse characteristic vectors corresponding to the associated articles respectively according to the article characteristics, inputting the sparse characteristic vectors corresponding to the associated articles respectively into an Embedding layer (Embedding), processing the sparse characteristic vectors through the Embedding layer to obtain dense characteristic vectors, and inputting the dense characteristic vectors corresponding to all the associated articles into a Mean pooling layer (Mean pooling) to obtain the behavior characteristic vectors;
Step 3: the user characteristic vector and the behavior characteristic vector are input into a splice layer Concat together, the user characteristic vector and the behavior characteristic vector are spliced through the Concat layer, and then the splicing result is input into a single-layer DNN network to obtain the user vector;
step 4: acquiring Item features (Item features) corresponding to the items to be recommended respectively, encoding the Item features to obtain sparse feature vectors, inputting the sparse feature vectors into an Embedding layer (Embedding), obtaining dense feature vectors corresponding to the items to be recommended, inputting the dense feature vectors into a SENET network for dynamic weighting to obtain Item feature vectors matched with the items to be recommended, and finally inputting the Item feature vectors into a DNN network to obtain Item vectors matched with the items to be recommended;
step 5: and calculating cosine similarity (cosin) between the user vector and each item vector, determining the correlation between the target user and the item to be recommended according to the cosine similarity, and recommending the target item to the target user according to the correlation.
Fig. 4 is a schematic structural diagram of a dual-tower model recommendation device based on behavior sequence and weight sharing according to a fourth embodiment of the present invention, where, as shown in fig. 4, the device includes: a user feature determination module 410, a behavioral feature determination module 420, an item vector determination module 430, and a target item determination module 440.
The user feature determining module 410 is configured to obtain at least one user feature corresponding to a target user, and process each user feature through a compression activation network SENet to obtain a user feature vector matched with the target user;
the behavior feature determining module 420 is configured to obtain at least one associated article corresponding to the target user according to the behavior sequence corresponding to the target user, and determine a behavior feature vector matched with the target user according to each associated article;
the item vector determining module 430 is configured to obtain item features corresponding to a plurality of items to be recommended respectively, and process each item feature through a SENet network to obtain an item vector matched with each item to be recommended;
the target item determining module 440 is configured to determine a target item from the plurality of items to be recommended according to the user feature vector, the behavior feature vector, and the item vector, and recommend the target item to a target user.
According to the technical scheme provided by the embodiment of the invention, the user characteristics corresponding to the target user are obtained through obtaining at least one user characteristic, the user characteristics are processed through the SENET network to obtain the user characteristic vector matched with the target user, at least one associated article corresponding to the target user is obtained according to the behavior sequence corresponding to the target user, the behavior characteristic vector matched with the target user is determined according to the associated articles, the article characteristics respectively corresponding to a plurality of articles to be recommended are obtained, the article characteristics are processed through the SENET network to obtain the article vector matched with the articles to be recommended, the target article is determined in the articles to be recommended according to the user characteristic vector, the behavior characteristic vector and the article vector, and the target article is recommended to the target user.
On the basis of the above embodiment, the user characteristic determining module 410 includes:
the user characteristic coding unit is used for coding each user characteristic to obtain a sparse characteristic vector corresponding to the target user;
the sparse vector processing unit is used for inputting the sparse feature vector into an embedded layer, and processing the sparse feature vector through the embedded layer to obtain a dense feature vector corresponding to a target user;
and the user feature vector determining unit is used for inputting the dense feature vector into a SENET network, and processing the dense feature vector through the SENET network to obtain a user feature vector matched with the target user.
The behavioral characteristics determination module 420 includes:
the associated article feature acquisition unit is used for acquiring article features corresponding to the associated articles respectively and generating sparse feature vectors corresponding to the associated articles respectively according to the article features;
the dense vector determining unit is used for inputting the sparse feature vectors corresponding to the associated articles respectively into the embedding layer, and processing the sparse feature vectors through the embedding layer to obtain dense feature vectors corresponding to the associated articles;
And the averaging unit is used for carrying out averaging processing on the dense feature vectors corresponding to all the associated objects to obtain behavior feature vectors matched with the target user.
The item vector determination module 430 includes:
the article feature vector determining unit is used for inputting the dense feature vector corresponding to each article to be recommended into a SENet network, and processing the dense feature vector through the SENet network to obtain an article feature vector matched with each article to be recommended;
and the item vector generation unit is used for inputting the item feature vectors into the dynamic neural network DNN to obtain item vectors matched with the items to be recommended.
The target item determination module 440 includes:
the user vector determining unit is used for determining a user vector corresponding to the target user according to the user characteristic vector and the behavior characteristic vector;
the relevance determining unit is used for determining the relevance between the target user and each article to be recommended according to the user vector and each article vector;
the target article screening unit is used for determining target articles from the plurality of articles to be recommended according to the correlation degree;
the vector splicing unit is used for splicing the user characteristic vector and the behavior characteristic vector, and inputting a splicing result into a DNN network to obtain a user vector corresponding to a target user;
And the similarity calculation unit is used for calculating cosine similarity between the user vector and each article vector and determining the correlation between the target user and each article to be recommended according to the cosine similarity.
The device can execute the method provided by all the embodiments of the invention, and has the corresponding functional modules and beneficial effects of executing the method. Technical details not described in detail in the embodiments of the present invention can be found in the methods provided in all the foregoing embodiments of the present invention.
Fig. 5 shows a schematic diagram of the structure of an electronic device 10 that may be used to implement an embodiment of the invention. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. Electronic equipment may also represent various forms of mobile devices, such as personal digital processing, cellular telephones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the inventions described and/or claimed herein.
As shown in fig. 5, the electronic device 10 includes at least one processor 11, and a memory, such as a Read Only Memory (ROM) 12, a Random Access Memory (RAM) 13, etc., communicatively connected to the at least one processor 11, in which the memory stores a computer program executable by the at least one processor, and the processor 11 may perform various appropriate actions and processes according to the computer program stored in the Read Only Memory (ROM) 12 or the computer program loaded from the storage unit 18 into the Random Access Memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 may also be stored. The processor 11, the ROM 12 and the RAM 13 are connected to each other via a bus 14. An input/output (I/O) interface 15 is also connected to bus 14.
Various components in the electronic device 10 are connected to the I/O interface 15, including: an input unit 16 such as a keyboard, a mouse, etc.; an output unit 17 such as various types of displays, speakers, and the like; a storage unit 18 such as a magnetic disk, an optical disk, or the like; and a communication unit 19 such as a network card, modem, wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information/data with other devices via a computer network, such as the internet, and/or various telecommunication networks.
The processor 11 may be a variety of general and/or special purpose processing components having processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various processors running machine learning model algorithms, digital Signal Processors (DSPs), and any suitable processor, controller, microcontroller, etc. The processor 11 performs the various methods and processes described above, such as a two-tower model recommendation method based on behavioral sequences and weight sharing.
In some embodiments, the dual tower model recommendation method based on behavior sequence and weight sharing may be implemented as a computer program tangibly embodied on a computer readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and/or installed onto the electronic device 10 via the ROM 12 and/or the communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the above-described dual tower model recommendation method based on behavior sequences and weight sharing may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform a dual tower model recommendation method based on behavior sequence and weight sharing in any other suitable way (e.g., by means of firmware).
Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuit systems, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), systems On Chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs, the one or more computer programs may be executed and/or interpreted on a programmable system including at least one programmable processor, which may be a special purpose or general-purpose programmable processor, that may receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device.
A computer program for carrying out methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions/acts specified in the flowchart and/or block diagram block or blocks to be implemented. The computer program may execute entirely on the machine, partly on the machine, as a stand-alone software package, partly on the machine and partly on a remote machine or entirely on the remote machine or server.
In the context of the present invention, a computer-readable storage medium may be a tangible medium that can contain, or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer readable storage medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. Alternatively, the computer readable storage medium may be a machine readable signal medium. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
To provide for interaction with a user, the systems and techniques described here can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) through which a user can provide input to the electronic device. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic input, speech input, or tactile input.
The systems and techniques described here can be implemented in a computing system that includes a background component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such background, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), blockchain networks, and the internet.
The computing system may include clients and servers. The client and server are typically remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also called a cloud computing server or a cloud host, and is a host product in a cloud computing service system, so that the defects of high management difficulty and weak service expansibility in the traditional physical hosts and VPS service are overcome.
It should be appreciated that various forms of the flows shown above may be used to reorder, add, or delete steps. For example, the steps described in the present invention may be performed in parallel, sequentially, or in a different order, so long as the desired results of the technical solution of the present invention are achieved, and the present invention is not limited herein.
The above embodiments do not limit the scope of the present invention. It will be apparent to those skilled in the art that various modifications, combinations, sub-combinations and alternatives are possible, depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of the present invention.
Claims (10)
1. A dual-tower model recommendation method based on behavior sequence and weight sharing, the method comprising:
acquiring at least one user characteristic corresponding to a target user, and processing each user characteristic through a compression activation network SENet to obtain a user characteristic vector matched with the target user;
according to the behavior sequence corresponding to the target user, at least one associated article corresponding to the target user is obtained, and according to each associated article, a behavior feature vector matched with the target user is determined;
Acquiring article characteristics corresponding to a plurality of articles to be recommended respectively, and processing the article characteristics through a SENet network to obtain article vectors matched with the articles to be recommended;
and determining a target object in the plurality of objects to be recommended according to the user feature vector, the behavior feature vector and the object vector, and recommending the target object to a target user.
2. The method of claim 1, wherein processing each of the user features by the compressed activation network SENet to obtain a user feature vector matching the target user comprises:
coding each user characteristic to obtain a sparse characteristic vector corresponding to a target user;
inputting the sparse feature vector into an embedded layer, and processing the sparse feature vector through the embedded layer to obtain a dense feature vector corresponding to a target user;
and inputting the dense feature vector into a SENET network, and processing the dense feature vector through the SENET network to obtain a user feature vector matched with the target user.
3. The method of claim 1, wherein determining a behavioral characteristic vector matching the target user based on each of the associated items comprises:
Acquiring article characteristics corresponding to each associated article respectively, and generating sparse characteristic vectors corresponding to each associated article respectively according to the article characteristics;
the sparse feature vectors corresponding to the associated articles are input into an embedding layer, and the sparse feature vectors are processed through the embedding layer to obtain dense feature vectors corresponding to the associated articles;
and carrying out mean value pooling treatment on the dense feature vectors corresponding to all the associated objects to obtain a behavior feature vector matched with the target user.
4. The method of claim 1, wherein processing each item feature through a SENet network to obtain an item vector matching each item to be recommended comprises:
inputting the dense feature vectors corresponding to the articles to be recommended into a SENet network, and processing the dense feature vectors through the SENet network to obtain article feature vectors matched with the articles to be recommended;
and inputting the item feature vectors into a dynamic neural network DNN to obtain item vectors matched with the items to be recommended.
5. The method of claim 1, wherein determining a target item from among a plurality of items to be recommended based on the user feature vector, behavior feature vector, and item vector comprises:
Determining a user vector corresponding to the target user according to the user feature vector and the behavior feature vector;
according to the user vector and each item vector, determining the correlation degree between the target user and each item to be recommended;
and determining the target object in the plurality of objects to be recommended according to the correlation.
6. The method of claim 5, wherein determining the user vector corresponding to the target user based on the user feature vector and the behavior feature vector comprises:
and splicing the user characteristic vector and the behavior characteristic vector, and inputting a splicing result into a DNN network to obtain a user vector corresponding to the target user.
7. The method of claim 5, wherein determining a relevance between a target user and each item to be recommended based on the user vector and each item vector comprises:
and calculating cosine similarity between the user vector and each article vector, and determining the correlation between the target user and each article to be recommended according to the cosine similarity.
8. A dual tower model recommendation device based on behavioral sequence and weight sharing, the device comprising:
The user characteristic determining module is used for acquiring at least one user characteristic corresponding to the target user, and processing each user characteristic through the compression activation network SENet to obtain a user characteristic vector matched with the target user;
the behavior feature determining module is used for acquiring at least one associated article corresponding to the target user according to the behavior sequence corresponding to the target user, and determining a behavior feature vector matched with the target user according to each associated article;
the article vector determining module is used for acquiring article characteristics corresponding to a plurality of articles to be recommended respectively, and processing the article characteristics through a SENet network to obtain article vectors matched with the articles to be recommended;
and the target article determining module is used for determining a target article in a plurality of articles to be recommended according to the user characteristic vector, the behavior characteristic vector and the article vector and recommending the target article to a target user.
9. An electronic device, the electronic device comprising:
at least one processor; and
a memory communicatively coupled to the at least one processor; wherein,,
the memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the behavior sequence and weight sharing based dual tower model recommendation method of any of claims 1-7.
10. A computer readable storage medium, characterized in that the computer readable storage medium stores computer instructions for causing a processor to implement the behavior sequence and weight sharing based two-tower model recommendation method according to any one of claims 1-7 when executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310136544.XA CN116308634A (en) | 2023-02-20 | 2023-02-20 | Double-tower model recommendation method and device based on behavior sequence and weight sharing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310136544.XA CN116308634A (en) | 2023-02-20 | 2023-02-20 | Double-tower model recommendation method and device based on behavior sequence and weight sharing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116308634A true CN116308634A (en) | 2023-06-23 |
Family
ID=86791596
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310136544.XA Pending CN116308634A (en) | 2023-02-20 | 2023-02-20 | Double-tower model recommendation method and device based on behavior sequence and weight sharing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116308634A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117408786A (en) * | 2023-12-11 | 2024-01-16 | 深圳须弥云图空间科技有限公司 | Article recommendation method and device based on gating mechanism |
-
2023
- 2023-02-20 CN CN202310136544.XA patent/CN116308634A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117408786A (en) * | 2023-12-11 | 2024-01-16 | 深圳须弥云图空间科技有限公司 | Article recommendation method and device based on gating mechanism |
| CN117408786B (en) * | 2023-12-11 | 2024-04-16 | 深圳须弥云图空间科技有限公司 | Article recommendation method and device based on gating mechanism |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112784778B (en) | Method, apparatus, device and medium for generating model and identifying age and sex | |
| CN112085565A (en) | Deep learning-based information recommendation method, device, equipment and storage medium | |
| CN113360711B (en) | Model training and executing method, device, equipment and medium for video understanding task | |
| CN113705628B (en) | Determination method and device of pre-training model, electronic equipment and storage medium | |
| CN113190702B (en) | Method and device for generating information | |
| CN113407851A (en) | Method, device, equipment and medium for determining recommendation information based on double-tower model | |
| CN114821063A (en) | Semantic segmentation model generation method and device and image processing method | |
| CN116308634A (en) | Double-tower model recommendation method and device based on behavior sequence and weight sharing | |
| CN114240555A (en) | Click rate prediction model training method and device and click rate prediction method and device | |
| CN113205189B (en) | Method for training prediction model, prediction method and device | |
| CN113592593A (en) | Training and application method, device, equipment and storage medium of sequence recommendation model | |
| CN114973333B (en) | Character interaction detection method, device, equipment and storage medium | |
| CN115169489B (en) | Data retrieval method, device, equipment and storage medium | |
| CN113656689B (en) | Model generation method and network information pushing method | |
| CN113255824B (en) | Method and apparatus for training classification model and data classification | |
| CN114662688A (en) | Model training method, data processing method, device, electronic device and medium | |
| CN115619142A (en) | Method, device, equipment and computer readable medium for matching data | |
| CN113239215B (en) | Classification method and device for multimedia resources, electronic equipment and storage medium | |
| CN114707638A (en) | Model training method, model training device, object recognition method, object recognition device, object recognition medium and product | |
| CN113989152A (en) | Image enhancement method, device, equipment and storage medium | |
| CN113010782A (en) | Demand amount acquisition method and device, electronic equipment and computer readable medium | |
| CN114998600B (en) | Image processing method, training method, device, equipment and medium for model | |
| CN116911913B (en) | Method and device for predicting interaction result | |
| CN114066278B (en) | Method, apparatus, medium, and program product for evaluating article recall | |
| CN113570432B (en) | Resource recommendation determining method, device, equipment and 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 |