CN114722269A - Article recommendation method and device based on graph neural network and storage medium - Google Patents

Article recommendation method and device based on graph neural network and storage medium Download PDF

Info

Publication number
CN114722269A
CN114722269A CN202210219700.4A CN202210219700A CN114722269A CN 114722269 A CN114722269 A CN 114722269A CN 202210219700 A CN202210219700 A CN 202210219700A CN 114722269 A CN114722269 A CN 114722269A
Authority
CN
China
Prior art keywords
user
article
information
layer
apegnn
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
Application number
CN202210219700.4A
Other languages
Chinese (zh)
Inventor
唐杰
张丹
朱一凡
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tsinghua University
Original Assignee
Tsinghua University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tsinghua University filed Critical Tsinghua University
Priority to CN202210219700.4A priority Critical patent/CN114722269A/en
Publication of CN114722269A publication Critical patent/CN114722269A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • 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
    • G06F16/9538—Presentation of query results
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • 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
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Business, Economics & Management (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Finance (AREA)
  • Evolutionary Computation (AREA)
  • Accounting & Taxation (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

In the article recommendation method, the article recommendation device and the storage medium based on the graph neural network, user information, article information and user-article interaction information are obtained, a directed network graph between a user and an article is constructed by using the user information, the article information and the user-article interaction information, an ApeGNN model based on the graph neural network is constructed, and the ApeGNN model is trained by using the directed network graph to obtain a target ApeGNN model. The method comprises the steps of obtaining information of an article to be recommended of each user, obtaining a prediction score of the article to be recommended by using a target ApeGNN model, determining a sorting result according to the prediction score, and outputting recommended article information corresponding to each user according to the sorting result. According to the method and the device, semantic differences of the user and the article in each layer are considered, so that the article recommendation accuracy is improved.

Description

Article recommendation method and device based on graph neural network and storage medium
Technical Field
The present application relates to the field of data recommendation technologies, and in particular, to an article recommendation method and apparatus based on a graph neural network, and a storage medium.
Background
With the rapid development of e-commerce and social media platforms, recommendation systems have become indispensable tools for enterprises to improve profits. At present, the graph neural network has achieved symbolic success on the recommendation system, and particularly, the graph neural network aggregates neighbor information on a graph structure through a message-passing process.
In the related art, the recommended models based on the graph neural network include the following:
the method comprises the following steps: and applying information propagation operation on the user-item bipartite graph, and recommending the possibly interested items for the user through the historical interaction records of the user.
The second method comprises the following steps: on the user-item bipartite graph, based on the recommendation model of the graph neural network, the sum-of-embedding average of each layer is taken as the characterization of the last layer in the pooling stage.
However, in method one, the user-item bipartite graph is a special type of graph, and the edges on the bipartite graph only have interaction information between the user and the item, and no edges exist between the user and the user or between the item and the item. In the second method, the semantic difference between the user and the article is ignored in the aggregation stage, and the importance of embedding different layers is not distinguished, so that the article recommendation accuracy is low.
Disclosure of Invention
The application provides an article recommendation method, an article recommendation device and a storage medium based on a graph neural network, which are used for at least solving the technical problem that the article recommendation accuracy is low due to the fact that semantic differences between users and articles are ignored in the related technology.
An embodiment of a first aspect of the present application provides an article recommendation method based on a graph neural network, including:
acquiring user information, article information and user-article interaction information;
constructing a directed network graph between the user and the object by using the user information, the object information and the user-object interaction information;
an ApeGNN model based on a graph neural network is built, and the ApeGNN model is trained by utilizing the directed network graph to obtain a target ApeGNN model;
acquiring information of articles to be recommended of each user;
and obtaining the prediction score of the item to be recommended by using the target ApeGNN model, determining an ordering result according to the prediction score, and outputting the recommended item information corresponding to each user according to the ordering result.
The embodiment of the second aspect of the present application provides an article recommendation device based on a graph neural network, including:
the acquisition module is used for acquiring user information, article information and user-article interaction information;
the building module is used for building a directed network graph between the user and the article by utilizing the user information, the article information and the user-article interaction information;
the training module is further used for constructing an ApeGNN model based on a neural network of a graph, and training the ApeGNN model by using the directed network graph to obtain a target ApeGNN model;
the acquisition module is also used for acquiring the information of the articles to be recommended by each user;
and the output module is used for obtaining the prediction scores of the to-be-recommended articles by using the target ApeGNN model, determining an ordering result according to the prediction scores, and outputting the recommended article information corresponding to each user according to the ordering result.
A non-transitory computer-readable storage medium according to an embodiment of a third aspect of the present application, where the non-transitory computer-readable storage medium stores a computer program; which when executed by a processor implements the method as shown in the first aspect above.
A computer device according to an embodiment of a fourth aspect of the present application includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the method according to the first aspect is implemented.
The technical scheme provided by the embodiment of the application at least has the following beneficial effects:
in the article recommendation method, the article recommendation device and the storage medium based on the graph neural network, user information, article information and user-article interaction information are obtained, a directed network graph between a user and an article is constructed by using the user information, the article information and the user-article interaction information, an ApeGNN model based on the graph neural network is constructed, the ApeGNN model is trained by using the directed network graph to obtain a target ApeGNN model, article information to be recommended of each user is obtained, a prediction score of the article to be recommended is obtained by using the target ApeGNN model, a sorting result is determined according to the prediction score, and recommended article information corresponding to each user is output according to the sorting result. According to the method and the device, the user and the article are taken as different types in the aggregation stage, and the graph convolution in the self-adaptive pooling layer configures different weights for different layers, so that the semantic difference of the user and the article in each layer is considered, and the article recommendation accuracy is improved.
Additional aspects and advantages of the present application will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the present application.
Drawings
The above and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
fig. 1 is a schematic flowchart of an item recommendation method based on a graph neural network according to an embodiment of the present application;
fig. 2 is a schematic structural diagram of an article recommendation device based on a graph neural network according to an embodiment of the present application.
Detailed Description
Reference will now be made in detail to embodiments of the present application, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the accompanying drawings are illustrative and intended to explain the present application and should not be construed as limiting the present application.
An article recommendation method and apparatus based on a graph neural network according to an embodiment of the present application are described below with reference to the drawings.
Example one
Fig. 1 is a schematic flowchart of an item recommendation method based on a graph neural network according to an embodiment of the present application, and as shown in fig. 1, the method may include:
step 101, obtaining user information, article information and user-article interaction information.
In an embodiment of the present invention, user information, item information, and user-item interaction information may be obtained from a transaction data set. And, the user information may include the user's ID, gender, age occupation, etc. for subsequent selection of the user's neighbor set using the user information. The item information may include an ID, type, price, etc. of the item for subsequent selection of a neighbor set of the item using the item information.
And, in the embodiment of the present invention, the user-item interaction information may include transaction information of the user and the item, and click browsing information of the user and the item.
And 102, constructing a directed network graph between the user and the article by utilizing the user information, the article information and the user-article interaction information.
For example, in the embodiment of the present invention, a directed network graph G between a user and an item is constructed as (U, V, R), where U represents a set of | U | user nodes, V represents a set of | V | item nodes, and R represents a set of interaction records between the user and the item, i.e., | R | edges. And uiRepresents a user node, vjRepresents an item node, r (u)ivj) Representing two nodes uiAnd vjEdge between if two nodes uiAnd vjWith interactive recording, r (u)ivj) Is 1, otherwise r (u)ivj) Is 0.
And 103, constructing an ApeGNN model based on the neural network of the graph, and training the ApeGNN model by using the directed network graph to obtain a target ApeGNN model.
Wherein, in an embodiment of the present invention, the ApeGNN model treats the user and the item as different types during the aggregation phase. And, in an embodiment of the present invention, the ApeGNN model may include, in order, an adaptive aggregation layer, an adaptive high-dimensional propagation layer, an adaptive pooling layer, and a prediction layer.
In an embodiment of the present invention, the method for training the ApeGNN model by using the directed net graph to obtain the target ApeGNN model may include the following steps:
step a, performing initialization embedding on all users and all articles according to the directed network graph to obtain an initialization embedding representation of the ith user
Figure BDA0003536610310000041
Initialized embedded representation of jth item
Figure BDA0003536610310000042
In the embodiment of the present invention, the embedded dimension is d, and m and n are the numbers of users and articles, respectively.
And b, respectively sequentially passing the user initialization embedded representation and the article initialization embedded representation through the self-adaptive aggregation layer, the self-adaptive high-dimensional propagation layer, the self-adaptive pooling layer and the prediction layer to obtain final prediction scores.
In an embodiment of the invention, when the user-initialized embedded representation and the article-initialized embedded representation pass through the adaptive aggregation layer, the adaptive aggregation layer embeds the representation to the user
Figure BDA0003536610310000051
And article embedded representation
Figure BDA0003536610310000052
Performing graph convolution operation by using embedded first aggregation function to obtain user embedded representation
Figure BDA0003536610310000053
And article embedded representation
Figure BDA0003536610310000054
And the first aggregation function is AGG1The first aggregation function is to aggregate the neighbor item sets of the users to obtain the representations of the users, and aggregate the neighbor user sets of the items to obtain the representations of the items, where the neighbor item sets of the users may be specifically item sets that have interacted with the users in the neighbor sets of the users, and the neighbor user sets of the items may be user sets that have interacted with the items in the neighbor sets of the items.
In particular, a user-embedded representation is obtained
Figure BDA0003536610310000055
And article embedded representation
Figure BDA0003536610310000056
Comprises the following steps:
Figure BDA0003536610310000057
Figure BDA0003536610310000058
wherein, N (u)i) Is user uiOf a neighbor set of thetauIs the only weight parameter that the user has t, L is the number of layers of the graph convolution network,
Figure BDA0003536610310000059
and
Figure BDA00035366103100000510
weights configured for the adaptive aggregation layer at the ith layer for the user and the item, respectively, e is a logarithmic function,
Figure BDA00035366103100000511
d is the diagonal node degree matrix and a is the adjacency matrix that does not contain self-joins.
Further, in the embodiment of the present invention, the adaptive high-dimensional propagation layer performs a graph convolution operation using the embedded second aggregation function and the result of the adaptive aggregation layer to search for the high-dimensional connection information for embedded propagation, wherein the second aggregation function is AGG2User uiAnd an article vjThe propagation embedding at the l-th layer is:
Figure BDA00035366103100000512
Figure BDA00035366103100000513
wherein the content of the first and second substances,
Figure BDA00035366103100000514
the number of convolution layers of the self-adaptive high-dimensional propagation layer and the self-adaptive aggregation layer is the same, and the convolution layers are L layers.
Further, in embodiments of the present invention, each layer in a graph convolution operation embeds semantic structure information that contains a different layer, based on which the embedding of each layer is handled separately in the pooling stage. Wherein, in the embodiment of the present disclosure, different weights are set for each layer to capture semantic information of each layer, so as to consider the importance of users and articles in different layers in the adaptive pooling stage.
Specifically, in the embodiment of the present disclosure, the adaptive pooling layer combines the results of the adaptive high-dimensional propagation layer to obtain the user uiAnd an article vjLast characterization
Figure BDA0003536610310000061
And
Figure BDA0003536610310000062
Figure BDA0003536610310000063
Figure BDA0003536610310000064
where β (u, l) is a weight configured for the l-th tier of users and β (ν, l) is a weight configured for the l-th tier of items.
Further, in embodiments of the present invention, the prediction layer will pass user uiIs characterized by
Figure BDA0003536610310000065
And an article vjIs characterized by
Figure BDA0003536610310000066
Inner product is carried out to obtain the final prediction fraction
Figure BDA0003536610310000067
And determining a sorting result according to the prediction score, and recommending the user according to the sorting result.
Wherein, in the embodiment of the invention, the user u is passediIs characterized by
Figure BDA0003536610310000068
And an article vjIs characterized by
Figure BDA0003536610310000069
Inner product is carried out to obtain the final prediction fraction
Figure BDA00035366103100000610
Can be as follows:
Figure BDA00035366103100000611
and, in one embodiment of the invention, the prediction scores may be ranked from high to low, and items ranked top (e.g., top 10) may be recommended according to the ranking results. In another embodiment of the invention, the prediction scores may be ranked from low to high, and ranked (e.g., 10 th) items may be recommended based on the ranking.
And c, optimizing the parameters of the ApeGNN model by using a Bayes personalized ranking loss function until the value of the loss function is not reduced any more, and obtaining the target ApeGNN model.
In an embodiment of the present invention, the bayesian personalized ranking loss function may be:
Figure BDA00035366103100000612
wherein the content of the first and second substances,
Figure BDA00035366103100000613
are the training data in pairs of the training data,
Figure BDA00035366103100000614
is a collection of interaction records that are,
Figure BDA00035366103100000615
is a collection of non-interactive records and,
Figure BDA00035366103100000616
is sigmoid function, lambda controls L2Regularization, while L2Regularization may prevent overfitting during the training process.
And, in embodiments of the present invention, the user per-layer weights and the item per-layer weights in the adaptive aggregation layer in the ApeGNN model can be optimized using a bayesian personalized ranking loss function, and the user per-layer weights and the item per-layer weights in the adaptive pooling layer in the ApeGNN model can be optimized.
And step 104, acquiring information of the articles to be recommended of each user.
And 105, obtaining a prediction score of the item to be recommended by using the target ApeGNN model, determining an ordering result according to the prediction score, and outputting recommended item information corresponding to each user according to the ordering result.
The method for recommending the articles based on the graph neural network comprises the steps of obtaining user information, article information and user-article interaction information, constructing a directed network graph between a user and the articles by utilizing the user information, the article information and the user-article interaction information, constructing an ApeGNN model based on the graph neural network, training the ApeGNN model by utilizing the directed network graph to obtain a target ApeGNN model, obtaining the information of the articles to be recommended of each user, obtaining a prediction score of the articles to be recommended by utilizing the target ApeGNN model, determining a sorting result according to the prediction score, and outputting recommended article information corresponding to each user according to the sorting result. According to the method and the device, the user and the article are taken as different types in the aggregation stage, and the graph convolution in the self-adaptive pooling layer configures different weights for different layers, so that the semantic difference of the user and the article in each layer is considered, and the article recommendation accuracy is improved.
Example two
Further, fig. 2 is a schematic structural diagram of an article recommendation device based on a graph neural network according to an embodiment of the present application, and as shown in fig. 2, the article recommendation device may include:
an obtaining module 201, configured to obtain user information, item information, and user-item interaction information;
a building module 202, configured to build a directed network graph between a user and an article by using the user information, the article information, and the user-article interaction information;
the training module 203 is further configured to construct an ApeGNN model based on a graph neural network, and train the ApeGNN model by using a directed network graph to obtain a target ApeGNN model;
the obtaining module 201 is further configured to obtain information of an item to be recommended by each user;
the output module 204 is configured to obtain a prediction score of the to-be-recommended item by using the target ApeGNN model, determine an ordering result according to the prediction score, and output recommended item information corresponding to each user according to the ordering result.
To implement the above embodiments, the present disclosure also proposes a non-transitory computer-readable storage medium.
A non-transitory computer-readable storage medium provided by an embodiment of the present disclosure stores a computer program; the computer program, when executed by a processor, is capable of implementing the method as shown in fig. 1.
In order to implement the above embodiments, the present disclosure also provides a computer device.
The computer device provided by the embodiment of the disclosure comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor; the processor, when executing the program, is capable of implementing the method as shown in fig. 1.
In the article recommendation method, the article recommendation device and the storage medium based on the graph neural network, user information, article information and user-article interaction information are obtained, a directed network graph between a user and an article is constructed by using the user information, the article information and the user-article interaction information, an ApeGNN model based on the graph neural network is constructed, the ApeGNN model is trained by using the directed network graph to obtain a target ApeGNN model, article information to be recommended of each user is obtained, a prediction score of the article to be recommended is obtained by using the target ApeGNN model, a sorting result is determined according to the prediction score, and recommended article information corresponding to each user is output according to the sorting result. According to the method and the device, the user and the article are taken as different types in the aggregation stage, and the graph convolution in the self-adaptive pooling layer configures different weights for different layers, so that the semantic difference of the user and the article in each layer is considered, and the article recommendation accuracy is improved.
In the description herein, reference to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the application. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing steps of a custom logic function or process, and alternate implementations are included within the scope of the preferred embodiment of the present application in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the present application.
Although embodiments of the present application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present application, and that variations, modifications, substitutions and alterations may be made to the above embodiments by those of ordinary skill in the art within the scope of the present application.

Claims (10)

1. An item recommendation method based on a graph neural network is characterized by comprising the following steps:
acquiring user information, article information and user-article interaction information;
constructing a directed network graph between the user and the object by using the user information, the object information and the user-object interaction information;
an ApeGNN model based on a graph neural network is built, and the ApeGNN model is trained by utilizing the directed network graph to obtain a target ApeGNN model;
acquiring information of articles to be recommended of each user;
and obtaining the prediction score of the item to be recommended by using the target ApeGNN model, determining an ordering result according to the prediction score, and outputting the recommended item information corresponding to each user according to the ordering result.
2. The method of claim 1, wherein the ApeGNN model comprises, in order, an adaptive aggregation layer, an adaptive high-dimensional propagation layer, an adaptive pooling layer, and a prediction layer, and wherein training the ApeGNN model with the directed net graph to obtain a target ApeGNN model comprises:
according to the directed network graph, all users and all articles are initialized and embedded to obtain an initialized and embedded representation of the ith user
Figure FDA0003536610300000011
Initialized embedded representation of jth item
Figure FDA0003536610300000012
The embedded dimensionality is d, and m and n are the numbers of users and articles respectively;
respectively sequentially passing the user initialization embedded representation and the article initialization embedded representation through a self-adaptive aggregation layer, a self-adaptive high-dimensional propagation layer, a self-adaptive pooling layer and a prediction layer to obtain final prediction scores;
and optimizing the parameters of the ApeGNN model by using a Bayes personalized ranking loss function until the value of the loss function is not reduced any more, thereby obtaining the target ApeGNN model.
3. The method as recited in claim 2, wherein the operation of the adaptive aggregation layer comprises:
the adaptive aggregation layer embeds representations for users
Figure FDA0003536610300000013
And an article embedded representation
Figure FDA0003536610300000014
Performing graph convolution operation by using embedded first aggregation function to obtain user embedded representation
Figure FDA0003536610300000015
And article embedded representation
Figure FDA0003536610300000016
Wherein the first aggregation function is AGG1，
The user-embedded representation
Figure FDA0003536610300000021
And article embedded representation
Figure FDA0003536610300000022
Comprises the following steps:
Figure FDA0003536610300000023
Figure FDA0003536610300000024
wherein, N (u)i) Is user uiOf a neighbor set of thetauIs the only weight parameter that the user has t, L is the number of layers of the graph convolution network,
Figure FDA0003536610300000025
and
Figure FDA0003536610300000026
weights configured for the adaptive aggregation layer at the ith layer for the user and the item, respectively, e is a logarithmic function,
Figure FDA0003536610300000027
is a diagonal node degree matrixAnd A is an adjacency matrix that does not contain self-connection.
4. The method as recited in claim 2, wherein said operation of said adaptive high-dimensional propagation layer comprises:
the self-adaptive high-dimensional propagation layer performs graph convolution operation by using the embedded second aggregation function and the result of the self-adaptive aggregation layer to perform embedded propagation and exploration of high-dimensional connection information, wherein the second aggregation function is AGG2User uiAnd an article vjThe propagation embedding at the l-th layer is:
Figure FDA0003536610300000028
Figure FDA0003536610300000029
wherein the content of the first and second substances,
Figure FDA00035366103000000210
is a symmetric normalization term.
5. The method of claim 2, wherein the operation of the adaptive pooling layer comprises:
the self-adaptive pooling layer combines the results of the self-adaptive high-dimensional propagation layer to obtain a user uiAnd an article vjLast characterization
Figure FDA00035366103000000211
And
Figure FDA00035366103000000212
Figure FDA00035366103000000213
Figure FDA00035366103000000214
wherein β (u, l) is a weight configured for the l-th layer of users, and β (ν, l) is a weight configured for the l-th layer of articles.
6. The method of claim 2, wherein the operation of predicting the layer comprises:
predict that layer will pass user uiIs characterized by
Figure FDA0003536610300000031
And an article vjIs characterized by
Figure FDA0003536610300000032
Inner product is carried out to obtain the final prediction fraction
Figure FDA0003536610300000033
And determining a sorting result according to the prediction score, and recommending the user according to the sorting result.
7. The method of claim 2, wherein the bayesian personalized ranking loss function comprises:
Figure FDA0003536610300000034
wherein the content of the first and second substances,
Figure FDA0003536610300000035
are the training data in pairs of the training data,
Figure FDA0003536610300000036
is a collection of interaction records that are,
Figure FDA0003536610300000037
is a collection of non-interactive records and,
Figure FDA0003536610300000038
is sigmoid function, lambda controls L2Regularization while L2Regularization may prevent overfitting during the training process.
8. An article recommendation device based on a graph neural network is characterized by comprising the following modules:
the acquisition module is used for acquiring user information, article information and user-article interaction information;
the building module is used for building a directed network graph between the user and the article by utilizing the user information, the article information and the user-article interaction information;
the training module is further used for constructing an ApeGNN model based on a neural network of a graph, and training the ApeGNN model by using the directed network graph to obtain a target ApeGNN model;
the acquisition module is also used for acquiring the information of the articles to be recommended by each user;
and the output module is used for obtaining the prediction scores of the to-be-recommended articles by using the target ApeGNN model, determining an ordering result according to the prediction scores, and outputting the recommended article information corresponding to each user according to the ordering result.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method as claimed in any one of claims 1 to 7 when the program is executed by the processor.
10. A non-transitory computer-readable storage medium, on which a computer program is stored, which, when executed by a processor, implements the method of any one of claims 1-7.
CN202210219700.4A 2022-03-08 2022-03-08 Article recommendation method and device based on graph neural network and storage medium Pending CN114722269A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210219700.4A CN114722269A (en) 2022-03-08 2022-03-08 Article recommendation method and device based on graph neural network and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210219700.4A CN114722269A (en) 2022-03-08 2022-03-08 Article recommendation method and device based on graph neural network and storage medium

Publications (1)

Publication Number Publication Date
CN114722269A true CN114722269A (en) 2022-07-08

Family

ID=82238290

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210219700.4A Pending CN114722269A (en) 2022-03-08 2022-03-08 Article recommendation method and device based on graph neural network and storage medium

Country Status (1)

Country Link
CN (1) CN114722269A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115688907A (en) * 2022-12-30 2023-02-03 中国科学技术大学 Recommendation model training method based on graph propagation and recommendation method based on graph propagation
CN116541716A (en) * 2023-07-06 2023-08-04 深圳须弥云图空间科技有限公司 Recommendation model training method and device based on sequence diagram and hypergraph
CN117112915A (en) * 2023-10-24 2023-11-24 广州美术学院 Intelligent design method and system based on user characteristics and big data training

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115688907A (en) * 2022-12-30 2023-02-03 中国科学技术大学 Recommendation model training method based on graph propagation and recommendation method based on graph propagation
CN116541716A (en) * 2023-07-06 2023-08-04 深圳须弥云图空间科技有限公司 Recommendation model training method and device based on sequence diagram and hypergraph
CN116541716B (en) * 2023-07-06 2024-04-16 深圳须弥云图空间科技有限公司 Recommendation model training method and device based on sequence diagram and hypergraph
CN117112915A (en) * 2023-10-24 2023-11-24 广州美术学院 Intelligent design method and system based on user characteristics and big data training
CN117112915B (en) * 2023-10-24 2024-02-20 广州美术学院 Intelligent design method and system based on user characteristics and big data training

Similar Documents

Publication Publication Date Title
CN111681059B (en) Training method and device of behavior prediction model
US20220198289A1 (en) Recommendation model training method, selection probability prediction method, and apparatus
US10489688B2 (en) Personalized digital image aesthetics in a digital medium environment
Kiran et al. DNNRec: A novel deep learning based hybrid recommender system
US20210248651A1 (en) Recommendation model training method, recommendation method, apparatus, and computer-readable medium
CN114722269A (en) Article recommendation method and device based on graph neural network and storage medium
CN110717098B (en) Meta-path-based context-aware user modeling method and sequence recommendation method
US8156056B2 (en) Method and system of classifying, ranking and relating information based on weights of network links
EP4181026A1 (en) Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium
CN112184391B (en) Training method of recommendation model, medium, electronic equipment and recommendation model
CN107545471B (en) Big data intelligent recommendation method based on Gaussian mixture
Biswas et al. A hybrid recommender system for recommending smartphones to prospective customers
US20220253722A1 (en) Recommendation system with adaptive thresholds for neighborhood selection
CN112287166A (en) Movie recommendation method and system based on improved deep belief network
US20220414470A1 (en) Multi-Task Attention Based Recurrent Neural Networks for Efficient Representation Learning
Pujahari et al. Preference relation based collaborative filtering with graph aggregation for group recommender system
Shiue et al. Development of machine learning‐based real time scheduling systems: using ensemble based on wrapper feature selection approach
CN115905687A (en) Cold start-oriented recommendation system and method based on meta-learning graph neural network
CN108984551A (en) A kind of recommended method and system based on the multi-class soft cluster of joint
US20220253688A1 (en) Recommendation system with adaptive weighted baysian personalized ranking loss
Alamdari et al. Deep reinforcement learning in seat inventory control problem: an action generation approach
Kalina et al. Robust training of radial basis function neural networks
US20240070743A1 (en) Systems and methods for recommended sorting of search results for online searching
CN111414535B (en) Method and device for recommending target object to user
CN112581185B (en) Advertisement user gender and age estimation method, system, computer 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