CN113191838B - Shopping recommendation method and system based on heterogeneous graph neural network - Google Patents
Shopping recommendation method and system based on heterogeneous graph neural network Download PDFInfo
- Publication number
- CN113191838B CN113191838B CN202110383787.4A CN202110383787A CN113191838B CN 113191838 B CN113191838 B CN 113191838B CN 202110383787 A CN202110383787 A CN 202110383787A CN 113191838 B CN113191838 B CN 113191838B
- Authority
- CN
- China
- Prior art keywords
- commodity
- user
- behavior
- vector
- neural network
- 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.)
- Active
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 76
- 238000000034 method Methods 0.000 title claims abstract description 35
- 239000013598 vector Substances 0.000 claims abstract description 157
- 230000006399 behavior Effects 0.000 claims abstract description 110
- 238000012549 training Methods 0.000 claims abstract description 31
- 239000011159 matrix material Substances 0.000 claims description 23
- 239000010410 layer Substances 0.000 claims description 14
- 238000004590 computer program Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 230000007246 mechanism Effects 0.000 claims description 6
- 238000005096 rolling process Methods 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 239000002356 single layer Substances 0.000 claims description 3
- 230000003213 activating effect Effects 0.000 claims 4
- 229910000831 Steel Inorganic materials 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 239000010959 steel Substances 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000005065 mining 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
-
- 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
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0282—Rating or review of business operators or products
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Finance (AREA)
- Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Theoretical Computer Science (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- General Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Data Mining & Analysis (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Economics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a shopping recommendation method and a shopping recommendation system based on a heterogeneous graph neural network, wherein the method comprises the following steps: acquiring historical behavior records of a plurality of users for commodities, and dividing the historical behavior records into a plurality of data sets according to the behaviors of the users for the commodities; for each behavior, generating initial user, commodity and commodity-to-commodity behavior relation feature vectors according to the corresponding data sets respectively; training by adopting a graph convolution neural network to obtain final users, commodities and characteristic vectors of the commodity behavior relation of the users; splicing the final user and the commodity feature vector to obtain a user-commodity feature vector; and obtaining the scoring prediction of the user for each commodity according to the user-commodity characteristic vector and the commodity behavior relation characteristic vector of the user. According to the method, a heterogeneous graph is formed by different users, commodities and the commodity behavior relation of the users, the link prediction of the relation between the users and the commodities is performed based on the graph convolution neural network, so that richer information can be introduced, and the recommendation precision is higher.
Description
Technical Field
The invention belongs to the technical field of personalized data recommendation, and particularly relates to a shopping recommendation method and system based on a heterogeneous graph neural network.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
Compared with the traditional shopping website, the types of commodities, the number of the commodities and other dimensional information on the platform are remarkably increased, and therefore a user cannot quickly and accurately find the commodities suitable for the user from one e-commerce website. In order to solve the problem, the recommendation system plays an important role gradually in the application process of the e-commerce website, and the personalized recommendation algorithm behind the recommendation system also becomes a research hot spot.
Traditional recommendation algorithms such as collaborative filtering algorithms, factorization machine algorithms and the like often cannot mine deep interests of users, and cold start problems caused by new users or new commodities are not well solved. The combination of deep learning and recommendation systems can generate better effect by means of the strong characterization capability of the neural network, but the existing deep learning algorithm models such as PNN, NCF and the like only use multi-layer neural network to perform deep crossing of project features in a plurality of modes, although the combination generates new feature information, the expression capability of the model is stronger than that of the traditional method, no more other types of information are introduced, no further research on the type of operation in the model is performed, for example, in the shopping scene, the user A purchases the book of how steel is obtained, and the existing method considers the features of commodities or users more, thus generating great limitation. That is, the existing recommendation algorithm still has the problems of inaccurate recommendation, weak generalization capability, incapability of mining deep interests of users and the like.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides a shopping recommendation method and system based on a heterogeneous graph neural network. The characteristics of different users, the characteristics of different commodities and the characteristics of different kinds of relations between the users and the commodities form a heterogeneous graph, the link prediction of the relation between the users and the commodities is carried out by using a graph convolution neural network mode, the commodity which is interested by the recommended users is finally realized, and richer information is introduced.
To achieve the above object, one or more embodiments of the present invention provide the following technical solutions:
a shopping recommendation method based on a heterogeneous graph neural network comprises the following steps:
acquiring historical behavior records of a plurality of users for commodities;
generating initial users, commodities and commodity behavior relation feature vectors of the users according to the historical behavior records;
training by adopting a graph convolution neural network to obtain final users, commodities and characteristic vectors of the commodity behavior relation of the users;
splicing the final user and the commodity feature vector to obtain a user-commodity feature vector;
and obtaining the scoring prediction of the user for each commodity according to the user-commodity characteristic vector and the commodity behavior relation characteristic vector of the user.
Further, based on the user identification information, the commodity identification information, and the user-to-commodity behavior relationship information, an initial user feature vector, an initial commodity feature vector, and an initial user-to-commodity behavior relationship feature vector are generated.
Further, the graph convolution neural network comprises a plurality of sub-graph convolution neural networks with the same layer number, the number of the sub-graph convolution neural networks is the same as the number of the behaviors of the user on the commodity, and the sub-graph convolution neural networks are respectively used for obtaining a third final feature vector based on two training of the initial user, the commodity and the commodity behavior relation feature vector of the user.
Further, an adjacency matrix of the behavior relationship between the user and the commodity is generated as a parameter of the graph convolution neural network according to the plurality of data sets.
Further, after the feature vector of the commodity behavior relation of the final user is obtained, an attention mechanism is applied to processing.
Further, according to the user-commodity feature vector and the commodity behavior relation feature vector of the user, obtaining the grading prediction of the user for each commodity comprises the following steps: and carrying out product operation on the user-commodity characteristic vector and the commodity behavior relation characteristic vector to obtain a scoring prediction matrix.
Further, the method further comprises: recommending commodities to a user according to the scoring prediction matrix comprises: and recommending the commodities with the highest scores to the users according to the commodity scores corresponding to each user according to the scoring prediction matrix.
One or more embodiments provide a shopping recommendation system based on a heterogram neural network, including:
the data acquisition module is configured to acquire historical behavior records of a plurality of users on commodities and divide the historical behavior records into a plurality of data sets according to the behaviors of the users on the commodities;
the initial vector training module is configured to generate initial user, commodity and commodity-to-commodity behavior relation feature vectors according to the corresponding data sets for each behavior;
the final vector training module is configured to train by adopting a graph convolution neural network to obtain final user, commodity and commodity behavior relation feature vectors of the user;
the recommendation score prediction module is configured to splice the final user and the commodity feature vector to obtain a user-commodity feature vector; and obtaining the scoring prediction of the user for each commodity according to the user-commodity characteristic vector and the commodity behavior relation characteristic vector of the user.
One or more embodiments provide an electronic device including a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the heterogeneous neural network-based shopping recommendation method when executing the program.
One or more embodiments provide a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the heterogeneous neural network-based shopping recommendation method.
The one or more of the above technical solutions have the following beneficial effects:
and constructing a heterogeneous graph by using the characteristics of different users, the characteristics of different commodities and the different kinds of relation characteristics between the users and the commodities, and predicting the link of the relation between the users and the commodities by using a graph convolution neural network mode, so that the commodities interested by the recommended users are finally realized, and the characteristic vectors can be fused with the information of adjacent nodes in the heterogeneous graph, so that the information of the adjacent nodes in the heterogeneous graph is richer.
The category of commodity behavior relations is taken into consideration by a user, and the memory capacity and the generalization capacity of the recommendation effect can be effectively improved by giving weights to different types of behavior relations by adopting a attention mechanism.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
FIG. 1 is a flowchart of a shopping recommendation method based on a heterogeneous graph neural network in one or more embodiments of the invention;
FIG. 2 is a schematic diagram of a process for scoring prediction based on a graph convolutional neural network in one or more embodiments of the invention.
Detailed Description
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
Embodiments of the invention and features of the embodiments may be combined with each other without conflict.
Example 1
The embodiment discloses a shopping recommendation method based on a heterogeneous graph neural network, which comprises the following steps as shown in fig. 1:
step 1: acquiring historical behavior records of a plurality of users for commodities, and dividing the behaviors of the commodities according to the users to obtain corresponding relationship data of the users and the commodities corresponding to the behaviors;
in this embodiment, the historical behavior of the user on the commodity is extracted through the log of the e-commerce platform. The actions for the commodity include: the purchasing behavior of the user, the behavior of adding goods to the shopping cart and the behavior of clicking the goods for viewing by the user.
According to the behavior classification, the data set corresponding to each behavior should be sorted in a form of a table, wherein the first column is the id of the user, and a plurality of columns are followed to represent the commodities that the user of the current id has performed, for example, the behavior record purchased by the user can be expressed as: the user id is followed by a plurality of item ids indicating that the specified user has operated on a plurality of items.
Step 2: generating initial users, commodities and commodity behavior relation feature vectors of the users according to the historical behavior records;
we apply the idea of Item2vec, train and generate the initial vector e of user feature by taking the user history behavior record as a sample u Commodity feature initial vector e v And an initial vector e of the characteristic of the commodity behavior relation of the user r . Specifically, an initial user feature vector, an initial merchandise feature vector, and an initial user-to-merchandise behavior relationship feature vector are generated based on the user identification information, the merchandise identification information, and the user-to-merchandise behavior relationship information.
Step 3: according to the initial characteristic vector of the commodity behavior relation of the user, commodity and user, training by adopting a graph convolution neural network to obtain the final characteristic vector of the commodity behavior relation of the user, commodity and user;
we apply the graph roll-up neural network GCN to the heterograms to make recommendations. And respectively applying different graph convolution neural networks to the purchase behavior of the user, the behavior of the user added to the shopping cart and the behavior of the user clicking commodity to view so as to extract the feature vector. The characteristic of the commodity forms a vector matrix E, the parameter is set as W, and the formula of the single-layer graph convolution neural network is defined as follows:
E (l) =σ(AE (l-1) W (l) )
wherein E is l And E is (l-1) Respectively represent the ith layer GCN andlayer i-1 GCN, σ is the Relu activation function
Firstly, we carry out graph rolling operation on the user commodity behavior data set obtained in the first step, and because the history data obtained in the first step is presented in a form of a table, and according to the conversion between the adjacency table representation of the graph and the adjacency matrix representation of the graph, we can easily obtain the adjacency matrix representation of the behavior relationship between the user and the commodity, the corresponding row and column are respectively the id vector representations of the corresponding user and the commodity, and the adjacency matrix representation obtained here is the parameter A in the graph rolling neural network.
The graph convolution neural network adopted in the embodiment comprises a plurality of sub-graph convolution neural networks with the same layer number. The number of the subgraph convolutional neural networks is the same as the number of the behaviors of the user on the commodity. In this embodiment, the graph convolution neural network includes 3 sub-graph convolution neural networks, and a third final feature vector is obtained based on training of two of the initial user, commodity and commodity behavior relation feature vectors of the user, respectively, and these result vectors are used as candidate sets in the personalized recall stage. < u1, u2, u3> represents an initial user feature vector, < v1, v2, v3> represents an initial commodity feature vector, respectively, < r1, r2, r3> represents an initial user-to-commodity behavior relationship feature vector, respectively, after obtaining 3 initial vectors, we train the remaining one vector with two of the vectors as parameters of the sub-GCN, respectively, and train for l times. As shown in fig. 2, the first subplot neural network is used for training based on < r1, r2, r3>, and < v1, v2, v3> to obtain final user feature vectors, and the second and third subplot neural networks are used for obtaining final commodity feature vectors and user-to-commodity behavior relation feature vectors, respectively. Taking training user vectors as an example, the training modes of other vectors are the same, and the formula of GCN for training user vectors is as follows:
wherein N is u And N v Representing the set of all direct neighbors of user u and commodity i in the heterograms respectively,representing the multiplication of matrix corresponding elements, σ is the Relu activation function. Note that here we are the final formula obtained after applying the parameter a obtained above to the GCN basic formula and finishing.
After each initial vector is trained for l times by the corresponding GCN network layer, a final vector representation is obtained, which is respectively:and->According to the principle of GCN, we can know that the vector at this time is fused with the information of the adjacent nodes in the heterogeneous graph, so that the candidate set vector at this time is more rich in the expression information.
Step 4: splicing the final user and the commodity feature vector to obtain a user-commodity feature vector, and simultaneously processing the commodity behavior relation feature vector by the user by applying an attention mechanism;
in step 3 we have trained separately representations of the user vector and the merchandise vector, which are independent, so here we first splice the user vector and the merchandise vector to get a new vector, denoted e uv 。
Since the final objective is to judge whether the user purchases the commodity or not as a recommended criterion for the user, the attention mechanism is applied to the commodity behavior relation feature vector to process, and for different kinds of relation vectors, different weights are given to the relation vector, the purchasing behavior is greater than the shopping cart adding behavior and greater than the checking behavior, and the weighted summation is carried out to obtain the final behavior relation vector representation, which is marked as e rf 。
Step 5: and obtaining the scoring prediction of the user for each commodity according to the user-commodity characteristic vector and the commodity behavior relation characteristic vector of the user.
After obtaining two final vectors, we perform product operation on the final vectors, the product operation result is used as the v score prediction of the user u on the appointed commodity, then we sort the final vectors according to the predicted scores, and the sorted front top k are recommended to the user.
The product operation formula is as follows:
wherein the method comprises the steps ofRepresenting the predictive score of user u for user v.
Example two
The embodiment aims to provide a shopping recommendation system based on a heterogeneous graphic neural network. The system comprises:
the data acquisition module is configured to acquire historical behavior records of a plurality of users for commodities;
the initial vector training module is configured to generate initial user, commodity and commodity behavior relation feature vectors of the user according to the historical behavior record;
the final vector training module is configured to train by adopting a graph convolution neural network to obtain final user, commodity and commodity behavior relation feature vectors of the user;
the recommendation score prediction module is configured to splice the final user and the commodity feature vector to obtain a user-commodity feature vector; and obtaining the scoring prediction of the user for each commodity according to the user-commodity characteristic vector and the commodity behavior relation characteristic vector of the user.
Example III
An object of the present embodiment is to provide an electronic apparatus.
An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the heterogeneous graphic neural network-based shopping recommendation method of embodiment one when executing the program.
Example IV
An object of the present embodiment is to provide a computer-readable storage medium.
A computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements a heterogeneous graphical neural network-based shopping recommendation method as described in embodiment one.
The steps involved in the second, third and fourth embodiments correspond to the first embodiment of the method, and the detailed description of the second embodiment refers to the relevant description of the first embodiment. The term "computer-readable storage medium" should be taken to include a single medium or multiple media including one or more sets of instructions; it should also be understood to include any medium capable of storing, encoding or carrying a set of instructions for execution by a processor and that cause the processor to perform any one of the methods of the present invention.
The technical scheme considers adding various information on the basis of considering the target characteristics. For example, user a purchases a book of how steel is refined, user B adds a shopping cart to a book of how steel is refined, and user B clicks and views a book of Kubernetes authoritative guide, which also considers the user's preference for behavior of the book of Kubernetes authoritative guide.
The technical scheme fully utilizes the data record generated by the user on the e-commerce platform, combines the characteristics of the user and the commodity entity, and also considers the relationship characteristics between the user and the commodity entity. More importantly, the influence of the category of the relationship on the preference of the user is considered, and although operations such as shopping cart adding, clicking and checking can be performed by the user to illustrate the preference of the user on the behavior of the commodity, the technical scheme only takes the operations as auxiliary factors, and the weight of the purchasing behavior is increased by adopting an attention mechanism to predict whether the user purchases the commodity, so that a conclusion on whether the commodity recommendation is performed for the user is obtained, and the recommendation effect is more accurate due to the consideration of the category of the relationship.
It will be appreciated by those skilled in the art that the modules or steps of the invention described above may be implemented by general-purpose computer means, alternatively they may be implemented by program code executable by computing means, whereby they may be stored in storage means for execution by computing means, or they may be made into individual integrated circuit modules separately, or a plurality of modules or steps in them may be made into a single integrated circuit module. The present invention is not limited to any specific combination of hardware and software.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
While the foregoing description of the embodiments of the present invention has been presented in conjunction with the drawings, it should be understood that it is not intended to limit the scope of the invention, but rather, it is intended to cover all modifications or variations within the scope of the invention as defined by the claims of the present invention.
Claims (10)
1. The shopping recommendation method based on the heterogeneous graph neural network is characterized by comprising the following steps of:
acquiring historical behavior records of a plurality of users for commodities;
generating initial users, commodities and commodity behavior relation feature vectors of the users according to the historical behavior records;
training by adopting a graph convolution neural network to obtain final users, commodities and characteristic vectors of the commodity behavior relation of the users; the method comprises the following steps:
applying the graph roll-up neural network GCN to the heterogeneous graph to recommend; extracting feature vectors by respectively applying different graph convolution neural networks to the purchase behavior of the user, the behavior of the user added to the shopping cart and the behavior of the user clicking commodity viewing; the characteristic of the commodity forms a vector matrix E, the parameter is set as W, and the formula of the single-layer graph convolution neural network is defined as follows:
wherein,and->Represents the i-th layer GCN and the i-1 th layer GCN,/respectively>Activating a function for Relu;
firstly, carrying out graph rolling operation on the user commodity behavior data set obtained in the first step, wherein the history data obtained in the first step is presented in a form of a table, and according to the conversion between the adjacency table representation of the graph and the adjacency matrix representation of the graph, adjacency matrix representations of the behavior relationship between the user and the commodity are obtained, the corresponding rows and columns are respectively the id vector representations of the corresponding user and the commodity, and the adjacency matrix representations obtained in the first step are the parameter A in the graph rolling neural network;
the adopted graph convolution neural network comprises a plurality of sub-graph convolution neural networks with the same layer number; the number of the sub-graph convolution neural networks is the same as the number of the behaviors of the users on the commodities, the sub-graph convolution neural networks comprise 3 sub-graph convolution neural networks, a third final feature vector is obtained based on training of two of initial user, commodity and commodity behavior relation feature vectors of the users, and the result vectors are used as candidate sets of the personalized recall stage;<u1,u2,u3>representing an initial user feature vector that is representative of the user,<v1,v2,v3>each representing an initial characteristic vector of the article,<r1,r2,r3>respectively representing the characteristic vectors of the initial user to commodity behavior relation, respectively training the rest vectors by using two of the initial vectors as parameters of sub GCN after obtaining 3 initial vectorsSecondary times; first oneSubgraph neural network for base on<r1,r2,r3>And (b)<v1,v2,v3>Training to obtain final user feature vectors, wherein the second and third feature vectors are respectively used for obtaining final commodity feature vectors and commodity behavior relation feature vectors of users; taking training user vectors as an example, the training modes of other vectors are the same, and the formula of GCN for training user vectors is as follows:
wherein,and->Representing the set of all direct neighbors of user u and commodity i in the heterograms, respectively, +.>Representing matrix multiplication, ++>Activating a function for Relu; note that here, the parameter a obtained above is applied to the GCN basic formula and the final formula obtained after finishing;
each initial vector is processed by the corresponding GCN network layerThe final vector representations obtained after the second training are respectively: />、/>And->According to the principle of GCNThe information of adjacent nodes in the heterogeneous graph is fused with the vector at the moment, so that the information of the selected vector representation is richer at the moment;
splicing the final user and the commodity feature vector to obtain a user-commodity feature vector;
and obtaining the scoring prediction of the user for each commodity according to the user-commodity characteristic vector and the commodity behavior relation characteristic vector of the user.
2. The shopping recommendation method based on the heterogeneous graphic neural network of claim 1, wherein the initial user feature vector, the initial commodity feature vector and the initial user-to-commodity behavior relationship feature vector are generated based on user identification information, commodity identification information and user-to-commodity behavior relationship information.
3. The shopping recommendation method based on the heterogeneous graph neural network according to claim 1, wherein the graph convolution neural network comprises a plurality of sub-graph convolution neural networks with the same layer number, the number of the sub-graph convolution neural networks is the same as the number of behaviors of the user on the commodity, and the sub-graph convolution neural networks are respectively used for training based on two of initial user, commodity and commodity-to-commodity behavior relation feature vectors to obtain a third final feature vector.
4. The shopping recommendation method based on the heterogeneous graphic neural network as claimed in claim 3, wherein based on the history of behavior records, an adjacency matrix of the behavior relationship between the user and the commodity is generated according to the behavior of the user on the commodity as a parameter of the graphic convolution neural network.
5. The shopping recommendation method based on the heterogeneous graphic neural network according to claim 1, wherein after obtaining the characteristic vector of the commodity behavior relation of the final user, an attention mechanism is applied for processing.
6. The heterogeneous neural network-based shopping recommendation method of claim 1, wherein obtaining a user's score prediction for each commodity according to the user-commodity feature vector and the user-commodity behavior relation feature vector comprises: and carrying out product operation on the user-commodity characteristic vector and the commodity behavior relation characteristic vector to obtain a scoring prediction matrix.
7. The heterogeneous neural network-based shopping recommendation method of claim 1, wherein the method further comprises: recommending commodities to a user according to the scoring prediction matrix comprises: and recommending the commodities with the highest scores to the users according to the commodity scores corresponding to each user according to the scoring prediction matrix.
8. A heterogeneous graphic neural network-based shopping recommendation system, comprising:
the data acquisition module is configured to acquire historical behavior records of a plurality of users for commodities;
the initial vector training module is configured to generate initial user, commodity and commodity behavior relation feature vectors of the user according to the historical behavior record;
the final vector training module is configured to train by adopting a graph convolution neural network to obtain final user, commodity and commodity behavior relation feature vectors of the user; the method comprises the following steps:
applying the graph roll-up neural network GCN to the heterogeneous graph to recommend; extracting feature vectors by respectively applying different graph convolution neural networks to the purchase behavior of the user, the behavior of the user added to the shopping cart and the behavior of the user clicking commodity viewing; the characteristic of the commodity forms a vector matrix E, the parameter is set as W, and the formula of the single-layer graph convolution neural network is defined as follows:
wherein,and->Represents the i-th layer GCN and the i-1 th layer GCN,/respectively>Activating a function for Relu;
firstly, carrying out graph rolling operation on the user commodity behavior data set obtained in the first step, wherein the history data obtained in the first step is presented in a form of a table, and according to the conversion between the adjacency table representation of the graph and the adjacency matrix representation of the graph, adjacency matrix representations of the behavior relationship between the user and the commodity are obtained, the corresponding rows and columns are respectively the id vector representations of the corresponding user and the commodity, and the adjacency matrix representations obtained in the first step are the parameter A in the graph rolling neural network;
the adopted graph convolution neural network comprises a plurality of sub-graph convolution neural networks with the same layer number; the number of the sub-graph convolution neural networks is the same as the number of the behaviors of the users on the commodities, the sub-graph convolution neural networks comprise 3 sub-graph convolution neural networks, a third final feature vector is obtained based on training of two of initial user, commodity and commodity behavior relation feature vectors of the users, and the result vectors are used as candidate sets of the personalized recall stage;<u1,u2,u3>representing an initial user feature vector that is representative of the user,<v1,v2,v3>each representing an initial characteristic vector of the article,<r1,r2,r3>respectively representing the characteristic vectors of the initial user to commodity behavior relation, respectively training the rest vectors by using two of the initial vectors as parameters of sub GCN after obtaining 3 initial vectorsSecondary times; the first sub-graph neural network is used for being based on<r1,r2,r3>And (b)<v1,v2,v3>Training to obtain final user feature vectors, wherein the second and third feature vectors are respectively used for obtaining final commodity feature vectors and commodity behavior relation feature vectors of users; taking training user vectors as an example, the training modes of other vectors are the same, and the formula of GCN for training user vectors is as follows:
wherein,and->Representing the set of all direct neighbors of user u and commodity i in the heterograms, respectively, +.>Representing matrix multiplication, ++>Activating a function for Relu; note that here, the parameter a obtained above is applied to the GCN basic formula and the final formula obtained after finishing;
each initial vector is processed by the corresponding GCN network layerThe final vector representations obtained after the second training are respectively: />、/>And->According to the principle of GCN, the vector at the moment is fused with the information of the adjacent nodes in the heterogeneous graph, so that the information of the selected vector representation is richer at the moment;
the recommendation score prediction module is configured to splice the final user and the commodity feature vector to obtain a user-commodity feature vector; and obtaining the scoring prediction of the user for each commodity according to the user-commodity characteristic vector and the commodity behavior relation characteristic vector of the user.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the heterogeneous graphical neural network-based shopping recommendation method of any of claims 1-7 when the program is executed by the processor.
10. A computer-readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements a shopping recommendation method based on a heterograph neural network as claimed in any one of claims 1-7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110383787.4A CN113191838B (en) | 2021-04-09 | 2021-04-09 | Shopping recommendation method and system based on heterogeneous graph neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110383787.4A CN113191838B (en) | 2021-04-09 | 2021-04-09 | Shopping recommendation method and system based on heterogeneous graph neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113191838A CN113191838A (en) | 2021-07-30 |
CN113191838B true CN113191838B (en) | 2024-01-30 |
Family
ID=76975325
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110383787.4A Active CN113191838B (en) | 2021-04-09 | 2021-04-09 | Shopping recommendation method and system based on heterogeneous graph neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113191838B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113254803B (en) * | 2021-06-24 | 2021-10-22 | 暨南大学 | Social recommendation method based on multi-feature heterogeneous graph neural network |
CN113643212B (en) * | 2021-08-27 | 2024-04-05 | 复旦大学 | Depth map noise reduction method based on map neural network |
CN114580794B (en) * | 2022-05-05 | 2022-07-22 | 腾讯科技(深圳)有限公司 | Data processing method, apparatus, program product, computer device and medium |
CN115080846A (en) * | 2022-05-30 | 2022-09-20 | 广州大学 | Scene information fused graph neural network clothing recommendation method |
CN117557318A (en) * | 2023-12-29 | 2024-02-13 | 青岛巨商汇网络科技有限公司 | Management intelligent analysis method and system based on virtual shopping |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110648163A (en) * | 2019-08-08 | 2020-01-03 | 中山大学 | Recommendation algorithm based on user comments |
WO2020149897A1 (en) * | 2019-01-17 | 2020-07-23 | Visa International Service Association | A deep learning model for learning program embeddings |
CN112380435A (en) * | 2020-11-16 | 2021-02-19 | 北京大学 | Literature recommendation method and recommendation system based on heterogeneous graph neural network |
-
2021
- 2021-04-09 CN CN202110383787.4A patent/CN113191838B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020149897A1 (en) * | 2019-01-17 | 2020-07-23 | Visa International Service Association | A deep learning model for learning program embeddings |
CN110648163A (en) * | 2019-08-08 | 2020-01-03 | 中山大学 | Recommendation algorithm based on user comments |
CN112380435A (en) * | 2020-11-16 | 2021-02-19 | 北京大学 | Literature recommendation method and recommendation system based on heterogeneous graph neural network |
Non-Patent Citations (1)
Title |
---|
Metapath-guided Heterogeneous Graph Neural Network for Intent Recommendation;shaohua Fan et.al.;《KDD "19: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining》;20190725;第2478-2486页 * |
Also Published As
Publication number | Publication date |
---|---|
CN113191838A (en) | 2021-07-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113191838B (en) | Shopping recommendation method and system based on heterogeneous graph neural network | |
CN109509054B (en) | Commodity recommendation method under mass data, electronic device and storage medium | |
CN110717098B (en) | Meta-path-based context-aware user modeling method and sequence recommendation method | |
CN110020128B (en) | Search result ordering method and device | |
CN106251174A (en) | Information recommendation method and device | |
CN105868847A (en) | Shopping behavior prediction method and device | |
CN107492008A (en) | Information recommendation method, device, server and computer-readable storage medium | |
CN110008973B (en) | Model training method, method and device for determining target user based on model | |
CN104239338A (en) | Information recommendation method and information recommendation device | |
CN108921602B (en) | User purchasing behavior prediction method based on integrated neural network | |
US20170186065A1 (en) | System and Method of Product Selection for Promotional Display | |
CN113268656A (en) | User recommendation method and device, electronic equipment and computer storage medium | |
Chen et al. | Common pitfalls in training and evaluating recommender systems | |
CN106202151A (en) | One is used for improving the multifarious method of personalized recommendation system | |
CN110598084A (en) | Object sorting method, commodity sorting device and electronic equipment | |
Seymen et al. | Customer churn prediction using deep learning | |
CN111680213B (en) | Information recommendation method, data processing method and device | |
CN112801803B (en) | Financial product recommendation method and device | |
CN108268519A (en) | A kind of method and apparatus of recommendation network object | |
CN110851694A (en) | Personalized recommendation system based on user memory network and tree structure depth model | |
CN115641179A (en) | Information pushing method and device and electronic equipment | |
CN116127188A (en) | Target feedback value determining method and device, electronic equipment and storage medium | |
CN112785391B (en) | Recommendation processing method and device, intelligent equipment and storage medium | |
CN116764236A (en) | Game prop recommending method, game prop recommending device, computer equipment and storage medium | |
Singh et al. | Using TOPSIS and modified TOPSIS methods for evaluating the competitive advantages of internet shopping malls |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |