CN113177159A - Binding recommendation method based on multichannel hypergraph neural network - Google Patents

Binding recommendation method based on multichannel hypergraph neural network Download PDF

Info

Publication number
CN113177159A
CN113177159A CN202110512230.6A CN202110512230A CN113177159A CN 113177159 A CN113177159 A CN 113177159A CN 202110512230 A CN202110512230 A CN 202110512230A CN 113177159 A CN113177159 A CN 113177159A
Authority
CN
China
Prior art keywords
user
bundle
article
hypergraph
package
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.)
Granted
Application number
CN202110512230.6A
Other languages
Chinese (zh)
Other versions
CN113177159B (en
Inventor
高跃
林浩杰
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 CN202110512230.6A priority Critical patent/CN113177159B/en
Publication of CN113177159A publication Critical patent/CN113177159A/en
Application granted granted Critical
Publication of CN113177159B publication Critical patent/CN113177159B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The application provides a binding recommendation method based on a multichannel hypergraph neural network, and relates to the technical field of recommendation systems, wherein the method comprises the following steps: acquiring a user-article-binding package interaction graph to construct a user hypergraph, an article hypergraph and a binding package hypergraph, acquiring node characteristics to initialize to obtain user characteristics, article characteristics and binding package characteristics, extracting the characteristics by using hypergraph convolution, obtaining new user characteristics, article characteristics and binding package characteristics, fusing to obtain corresponding user characteristic representation, article characteristic representation and binding package characteristic representation, and using an article pooling module to aggregate the article characteristic representations and add the object characteristic representations to the binding package characteristic representation to obtain final representation; and multiplying the user characteristic representation and the final representation point to obtain the preference degree of the user to the bundle package, and then performing bundle recommendation. The invention adopting the scheme solves the technical problems that the prior method does not explicitly distinguish and model the user, the article and the bundle package, and is still deficient when modeling high-order association.

Description

Binding recommendation method based on multichannel hypergraph neural network
Technical Field
The application relates to the technical field of recommendation systems, in particular to a binding recommendation method based on a multichannel hypergraph neural network and computer equipment.
Background
As an effective way to solve the information overload problem, recommendation systems have been widely applied in the fields of music platforms, short video platforms, and the like. Most of the existing recommendation system methods focus on how to better recommend a single item to a user, however, in part of recommendation scenarios, in addition to a single item, the recommendation system needs to recommend a bundle package containing multiple items to the user, i.e., a bundle recommendation. In bundle recommendations, we generally have dependency information for the item and bundle, interaction information for the user with the bundle, and interaction information for the user and the item. Based on this information, some work began using graph networks to model relationships between items, bundles, and users. However, existing approaches do not explicitly model users, items, and bundles differentially, and are still deficient when modeling high-order associations.
Disclosure of Invention
The present application is directed to solving, at least to some extent, one of the technical problems in the related art.
Therefore, the first purpose of the application is to provide a binding recommendation method based on a multichannel hypergraph neural network, which solves the technical problems that the existing method does not explicitly distinguish and model the user, the article and the binding package, and is still deficient when modeling high-order association, realizes the purpose of introducing a hypergraph structure to model high-order association information among different types of nodes, and simultaneously realizes the purposes of modeling by using information of multiple different interaction types in the task and explicitly distinguishing when modeling the user, the article and the binding package.
The second purpose of the application is to provide a binding recommendation system based on the multichannel hypergraph neural network.
A third object of the present application is to propose a computer device.
In order to achieve the above object, an embodiment of a first aspect of the present application provides a binding recommendation method based on a multi-channel hypergraph neural network, including: acquiring a user-item-bundle package interaction graph and corresponding node characteristics, wherein the user-item-bundle package interaction graph comprises subordinate information of items and a bundle package, interaction information of a user and the bundle package and interaction information of the user and the items;
constructing a user hypergraph, an item hypergraph and a bundle package hypergraph according to the user-item-bundle package interaction graph;
initializing the node characteristics to obtain user characteristics, article characteristics and bundle package characteristics;
carrying out feature extraction on the user features, the article features and the bundle package features and corresponding user hypergraphs, article hypergraphs and bundle package hypergraphs by using hypergraph convolution to obtain new user features, article features and bundle package features;
carrying out feature fusion on the new user features, the article features and the bundle package features to obtain corresponding user feature representation, article feature representation and bundle package feature representation;
aggregating the item representations using a bundle-guided item pooling module and adding the aggregated item representations to the bundle representation to obtain a final representation;
performing point multiplication on the user characteristic representation and the final representation to obtain the preference degree of the user to the bundle package;
and performing binding recommendation according to the preference degree of the binding package.
Optionally, in an embodiment of the present application, the building of the user hypergraph includes the following steps:
for the interactive information between the user and the articles, taking each article as a center, and connecting the user directly connected with the articles by using the super edges so as to generate a first group of super edges;
for the interactive information of the user and the binding packages, taking each binding package as a center, and connecting the user directly connected with the article by using the super edge so as to generate a second group of super edges;
fusing the first group of the super edges and the second group of the super edges by adopting a splicing fusion mode to generate a user super graph,
the construction of the article hypergraph comprises the following steps:
for the interactive information between the users and the articles, taking each user as a center, and connecting the articles directly connected with the users by using the super edges so as to generate a third group of super edges;
regarding the subordinate information of the user and the binding packages, connecting the objects in each binding package by using the super edges with each binding package as the center, thereby generating a fourth group of super edges;
fusing the third group of the super-edges and the fourth group of the super-edges by adopting a splicing fusion mode to generate an article super-image,
the construction of the bundle hypergraph comprises the following steps:
for the interactive information of the users and the binding packages, taking each user as a center, and connecting the binding packages directly connected with a certain article by using the super edges so as to generate a fifth group of super edges;
for the subordinate information of the articles and the binding packets, the binding packets directly connected with one article are connected by using the super edges, so that a sixth group of super edges is generated;
and fusing the fifth group of the super edges and the sixth group of the super edges by adopting a splicing fusion mode to generate the bundle package hypergraph.
Optionally, in an embodiment of the present application, the user features, the item features, the bundle package features, and the corresponding user hypergraph, item hypergraph, and bundle package hypergraph are subjected to feature extraction by using hypergraph convolution, so as to obtain new user features, item features, and bundle package features, which are expressed as:
Figure BDA0003060813940000021
Figure BDA0003060813940000022
Figure BDA0003060813940000031
wherein the content of the first and second substances,
Figure BDA0003060813940000032
and
Figure BDA0003060813940000033
respectively being the user characteristic and the article characteristicAnd the bundle characteristic;
Figure BDA0003060813940000034
Figure BDA0003060813940000035
and
Figure BDA0003060813940000036
the corresponding output characteristics;
Figure BDA0003060813940000037
and
Figure BDA0003060813940000038
leakly _ Relu is an activation function for the learnable parameters of the module, HuIn order to allow the user to go beyond the picture,
Figure BDA0003060813940000039
is a degree matrix of the user hypergraph nodes,
Figure BDA00030608139400000310
a matrix, W, which is the superedge of the user hypergraphuWeight matrix for the user hypergraph hyper-edge, HiThe method is characterized in that the method is a super map of an article,
Figure BDA00030608139400000311
is a degree matrix of the nodes of the commodity hypergraph,
Figure BDA00030608139400000312
is a matrix of the over-edges of the hypergraph of the article, WiWeight matrix for the hyper-edge of the item hypergraph, HbIn order to super-map the bundle package,
Figure BDA00030608139400000313
for the degree matrix of the bundle hypergraph nodes,
Figure BDA00030608139400000314
for binding the matrices of the super-edges of the package, WbA weight matrix that is the super edge of the bundle hypergraph.
Optionally, in an embodiment of the present application, features of users, items and bundles at different levels are fused, expressed as:
Figure BDA00030608139400000315
Figure BDA00030608139400000316
Figure BDA00030608139400000317
wherein u isjRepresents a user, ikRepresenting an article, bpIndicating a bundle, L being the number of network layers
Figure BDA00030608139400000318
As a feature of the user, it is,
Figure BDA00030608139400000319
is characterized by the characteristics of the article,
Figure BDA00030608139400000320
is a bundle feature.
Optionally, in one embodiment of the application, a bundle-guided item pooling module is constructed to pool item features within a bundle by:
Figure BDA00030608139400000321
wherein the content of the first and second substances,
Figure BDA00030608139400000322
Figure BDA00030608139400000323
and
Figure BDA00030608139400000324
to trainable the parameters, σ is an activation function, in particular a LeakyRelu activation function,
Figure BDA00030608139400000325
as a bundle package bpIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure BDA00030608139400000326
is an article ikIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure BDA00030608139400000327
is an article ilIs characterized by apkFor the weight coefficient of the pooling module, JpSet of items contained in bundle p, ilIs an article il，bpAs a bundle, bpAssembled by articles
Figure BDA00030608139400000328
Is composed of, and article ikIs a set
Figure BDA00030608139400000329
One of the articles;
optionally, in an embodiment of the present application, the user feature representation and the final representation are subjected to dot multiplication, specifically represented as:
Figure BDA00030608139400000330
wherein the content of the first and second substances,
Figure BDA00030608139400000331
in order to be represented by the characteristics of the user,
Figure BDA00030608139400000332
to be the final representation.
In order to achieve the above object, a second aspect of the present invention provides a binding recommendation system based on a multi-channel hypergraph neural network, including:
the first acquisition module is used for acquiring a user-article-bundle package interaction graph and corresponding node characteristics;
the building module is used for building a user hypergraph, an article hypergraph and a bundle hypergraph according to the user-article-bundle package interaction graph, and initializing according to the node characteristics to obtain user characteristics, article characteristics and bundle package characteristics;
the second acquisition module is used for extracting the user characteristics, the article characteristics and the bundle package characteristics and corresponding user hypergraph, article hypergraph and bundle package hypergraph by using hypergraph convolution to obtain new user characteristics, article characteristics and bundle package characteristics;
the fusion module is used for carrying out feature fusion on the new user features, the article features and the bundle package features to obtain corresponding user feature representation, article feature representation and bundle package feature representation;
a third obtaining module, configured to aggregate the item feature representations using the item pooling module, and add the aggregated item feature representations to the bundle feature representation to obtain a final representation;
and the determining module is used for obtaining the preference degree of the user to the bundle package by performing point multiplication on the user characteristic representation and the final representation.
Optionally, in an embodiment of the present application, training is performed on the system, and learning is performed by using bayesian personalized ranking, where the specific process is as follows:
constructing a training set triple set for training, wherein the triple set is represented as:
Figure BDA0003060813940000041
wherein
Figure BDA0003060813940000042
For recording the interactions of the user with the bundle package that have occurred in the training set, and
Figure BDA0003060813940000043
is the record of interaction between the user and the bundle that does not appear in the training set, u is the user, i is a bundle clicked by the user u in the training set,? j is a bundle that user u did not click on in the training set,
training by the following function model based on the triplet set:
Figure BDA0003060813940000044
where σ is the Sigmoid function, Θ is the trainable parameter of the model, λ is the coefficient of the $ L2 regular term, σ is the activation function,
Figure BDA0003060813940000045
to the extent user u prefers bundle i,
Figure BDA0003060813940000046
the preference of user u for bundle j.
In order to achieve the above object, an embodiment of a third aspect of the present invention provides a computer apparatus, including: the device comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, wherein the processor realizes the binding recommendation method based on the multichannel hypergraph neural network when executing the computer program.
According to the binding recommendation method based on the multichannel hypergraph neural network, the computer device and the non-transitory computer readable storage medium, a hypergraph structure is introduced to model high-order associated information among different types of nodes, meanwhile, the information of multiple different interaction types such as subordinate information of objects and binding packages, interaction information of users and objects and the like in the task are used for modeling, and the users, the objects and the binding packages are explicitly distinguished when being modeled.
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 foregoing 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 flowchart of a binding recommendation method based on a multi-channel hypergraph neural network according to an embodiment of the present application;
FIG. 2 is another flowchart of a binding recommendation method based on a multi-channel hypergraph 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 drawings are exemplary and intended to be used for explaining the present application and should not be construed as limiting the present application.
The following describes a binding recommendation method based on a multi-channel hypergraph neural network according to an embodiment of the present application with reference to the drawings.
Fig. 1 is a flowchart of a binding recommendation method based on a multi-channel hypergraph neural network according to an embodiment of the present application.
As shown in fig. 1, the binding recommendation method based on the multi-channel hypergraph neural network includes the following steps:
step 101, acquiring a user-item-bundle package interaction graph and corresponding node characteristics, wherein the user-item-bundle package interaction graph comprises subordinate information of items and bundle packages, interaction information of users and bundle packages and interaction information of users and items;
102, constructing a user hypergraph, an item hypergraph and a bundle package hypergraph according to the user-item-bundle package interaction graph;
step 103, initializing the node characteristics to obtain user characteristics, article characteristics and bundle package characteristics;
104, extracting the user characteristics, the article characteristics and the binding package characteristics and the corresponding user hypergraph, article hypergraph and binding package hypergraph by using hypergraph convolution to obtain new user characteristics, article characteristics and binding package characteristics;
step 105, carrying out feature fusion on the new user features, the article features and the bundle package features to obtain corresponding user feature representation, article feature representation and bundle package feature representation;
step 106, aggregating the item feature representations by using the bundle-guided item pooling module, and adding the aggregated item feature representations and the bundle feature representations to obtain a final representation;
step 107, the preference degree of the user to the bundle package is obtained by performing point multiplication on the user characteristic representation and the final representation;
and step 108, carrying out binding recommendation according to the preference degree of the binding package.
According to the binding recommendation method based on the multichannel hypergraph neural network, a user-article-binding package interaction graph and corresponding node characteristics are obtained, wherein the user-article-binding package interaction graph comprises subordinate information of articles and binding packages, interaction information of users and the binding packages and interaction information of the users and the articles; constructing a user hypergraph, an item hypergraph and a bundle package hypergraph according to the user-item-bundle package interaction graph; initializing the node characteristics to obtain user characteristics, article characteristics and bundle package characteristics; carrying out feature extraction on the user features, the article features and the bundle package features and corresponding user hypergraphs, article hypergraphs and bundle package hypergraphs by using hypergraph convolution to obtain new user features, article features and bundle package features; carrying out feature fusion on the new user features, the article features and the bundle package features to obtain corresponding user feature representation, article feature representation and bundle package feature representation; aggregating the item representations using a bundle-guided item pooling module and adding the aggregated item representations to the bundle representation to obtain a final representation; performing point multiplication on the user characteristic representation and the final representation to obtain the preference degree of the user to the bundle package; and performing binding recommendation according to the preference degree of the binding package. Therefore, the method and the device can solve the technical problems that the existing method does not perform explicit distinguishing modeling on the user, the article and the bundle package, and is still deficient in modeling high-order association, so that a hypergraph structure is introduced to model high-order association information among different types of nodes, meanwhile, information of multiple different interaction types in the task is used for modeling, and explicit distinguishing is performed when the user, the article and the bundle package are modeled.
Further, in the embodiment of the present application, a three-part graph including three nodes of a user, an article and a bundle and corresponding node characteristics are given. The graph structure contains dependency information for the item and the bundle, interaction information for the user with the bundle, and interaction information for the user with the item. By using
Figure BDA0003060813940000061
And
Figure BDA0003060813940000062
representing a user set, an item set, and a bundle set, where N, M and O represent a user number, an item number, and a bundle number, respectively. By using
Figure BDA0003060813940000063
And
Figure BDA0003060813940000064
respectively representing the interaction information of the user and the article, the interaction information of the user and the bundle package and the subordinate information of the article and the bundle package. Initialization of features for users, items and bundles
Figure BDA0003060813940000065
And
Figure BDA0003060813940000066
and for the user, constructing the hypergraph by utilizing the interaction information of the user and the article and the interaction information of the user and the bundle package.
The construction of the user hypergraph comprises the following steps:
for the interactive information between the user and the articles, taking each article as a center, and connecting the user directly connected with the articles by using the super edges so as to generate a first group of super edges;
for the interactive information of the user and the binding packages, taking each binding package as a center, and connecting the user directly connected with the article by using the super edge so as to generate a second group of super edges;
the first group of super edges and the second group of super edges are fused in a splicing fusion mode to generate a user super graph, which specifically comprises the following steps:
Hu,1＝P
Hu,2＝Q
Hu＝Hu,1|Hu,2
for the item, constructing a hypergraph by utilizing the interaction information of the user and the item and the subordinate information of the item and the bundle package.
The construction of the article hypergraph comprises the following steps:
for the interactive information between the users and the articles, taking each user as a center, and connecting the articles directly connected with the users by using the super edges so as to generate a third group of super edges;
regarding the subordinate information of the user and the binding packages, connecting the objects in each binding package by using the super edges with each binding package as the center, thereby generating a fourth group of super edges;
and fusing the third group of super edges and the fourth group of super edges by adopting a splicing fusion mode to generate an article super map, which specifically comprises the following steps:
Hi，1＝PT
Hi,2＝RT
Hi＝Hi,1|Hi,2
and for the bundle package, constructing a hypergraph by utilizing the interaction information of the user and the bundle package and the subordinate information of the object and the bundle package.
The construction of the bundle hypergraph comprises the following steps:
for the interactive information of the users and the binding packages, taking each user as a center, and connecting the binding packages directly connected with a certain article by using the super edges so as to generate a fifth group of super edges;
for the subordinate information of the object and the binding package, connecting the binding package directly connected with one object by using the super edge so as to generate a first group of super edges;
and fusing the fifth group of the super edges and the sixth group of the super edges by adopting a splicing fusion mode to generate a bundle package super graph, which specifically comprises the following steps:
Hb,1＝QT
Hb,2＝R
Hb＝Hb,1|Hb,2
further, in this embodiment of the present application, based on the node features input by the multi-channel hypergraph and the model, the hypergraph convolution is used to capture the high-order associated information, and the hypergraph convolution is used to perform feature extraction on the user features, the item features, and the bundle package features, so as to obtain new user features, item features, and bundle package features, which are expressed as:
Figure BDA0003060813940000071
Figure BDA0003060813940000072
Figure BDA0003060813940000073
wherein the content of the first and second substances,
Figure BDA0003060813940000074
and
Figure BDA0003060813940000075
the user characteristic, the item characteristic, and the bundle characteristic, respectively;
Figure BDA0003060813940000076
Figure BDA0003060813940000077
and
Figure BDA0003060813940000078
the corresponding output characteristics;
Figure BDA0003060813940000079
and
Figure BDA00030608139400000710
leakly _ Relu is an activation function for the learnable parameters of the module, HuIn order to allow the user to go beyond the picture,
Figure BDA00030608139400000711
is a degree matrix of the user hypergraph nodes,
Figure BDA00030608139400000712
a matrix, W, which is the superedge of the user hypergraphuWeight matrix for the user hypergraph hyper-edge, HiThe method is characterized in that the method is a super map of an article,
Figure BDA00030608139400000713
is a degree matrix of the nodes of the commodity hypergraph,
Figure BDA00030608139400000714
is a matrix of the over-edges of the hypergraph of the article, WiWeight matrix for the hyper-edge of the item hypergraph, HbIn order to super-map the bundle package,
Figure BDA00030608139400000715
for the degree matrix of the bundle hypergraph nodes,
Figure BDA00030608139400000716
for binding the matrices of the super-edges of the package, WbA weight matrix that is the super edge of the bundle hypergraph.
Further, in the embodiment of the present application, in order to perform model prediction, features of users, items and bundles at different levels are fused to obtain corresponding feature representations. Fusing the characteristics of the users, the items and the bundle packages at different levels, and expressing as follows:
Figure BDA0003060813940000081
Figure BDA0003060813940000082
Figure BDA0003060813940000083
wherein u isjRepresents a user, ikRepresenting an article, bpIndicating a bundle, L being the number of network layers
Figure BDA0003060813940000084
As a feature of the user, it is,
Figure BDA0003060813940000085
is characterized by the characteristics of the article,
Figure BDA0003060813940000086
is a bundle feature.
Further, in the embodiment of the present application, in order to better complement the characteristics of the bundle package with the characteristics of the items in the bundle package, a bundle package guided item pooling module is proposed, and the bundle package guided item pooling module is constructed to pool the characteristics of the items in the bundle package by:
Figure BDA0003060813940000087
wherein the content of the first and second substances,
Figure BDA0003060813940000088
and
Figure BDA0003060813940000089
to trainable the parameters, σ is an activation function, in particular a LeakyRelu activation function,
Figure BDA00030608139400000810
as a bundle package bpIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure BDA00030608139400000811
is an article ikIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure BDA00030608139400000812
is an article ilIs characterized by apkFor the weight coefficient of the pooling module, JpSet of items contained in bundle p, ilIs an article il，bpAs a bundle, bpAssembled by articles
Figure BDA00030608139400000813
Is composed of, and article ikIs a set
Figure BDA00030608139400000814
One of the articles.
Further, in the embodiments of the present application, the bundle-guided item pooling module is used to aggregate and add item characteristics to the bundle characteristics to obtain a final representation:
Figure BDA00030608139400000815
the preference degree of the user to the bundle package is obtained by performing point multiplication on the characteristics of the user and the characteristics of the bundle package, and the user characteristics and the characteristics of the new bundle package are specifically represented as follows:
Figure BDA00030608139400000816
wherein the content of the first and second substances,
Figure BDA00030608139400000817
in order to be represented by the characteristics of the user,
Figure BDA00030608139400000818
to be the final representation.
Fig. 2 is another flowchart of a binding recommendation method based on a multi-channel hypergraph neural network according to an embodiment of the present application.
As shown in fig. 2, the binding recommendation method based on the multi-channel hypergraph neural network includes: constructing a user hypergraph, an item hypergraph and a bundle package hypergraph based on the input user-item-bundle package interaction graph; based on the node characteristics input by the multichannel hypergraph and the model, performing high-order associated information capture by using hypergraph convolution; supplementing bundle features with a bundle-guided item pooling module; fusing the characteristics of the users, the articles and the bundle packages at different levels to obtain corresponding characteristic representations; and for the training of the model, Bayesian personalized sorting is adopted for learning.
In order to implement the above embodiments, the present invention further provides a binding recommendation system based on a multi-channel hypergraph neural network, including:
the first acquisition module is used for acquiring a user-article-bundle package interaction graph and corresponding node characteristics;
the building module is used for building a user hypergraph, an article hypergraph and a bundle hypergraph according to the user-article-bundle package interaction graph, and initializing according to the node characteristics to obtain user characteristics, article characteristics and bundle package characteristics;
the second acquisition module is used for extracting the user characteristics, the article characteristics and the bundle package characteristics and corresponding user hypergraph, article hypergraph and bundle package hypergraph by using hypergraph convolution to obtain new user characteristics, article characteristics and bundle package characteristics;
the fusion module is used for carrying out feature fusion on the new user features, the article features and the bundle package features to obtain corresponding user feature representation, article feature representation and bundle package feature representation;
a third obtaining module, configured to aggregate the item feature representations using the item pooling module, and add the aggregated item feature representations to the bundle feature representation to obtain a final representation;
and the determining module is used for obtaining the preference degree of the user to the bundle package by performing point multiplication on the user characteristic representation and the final representation.
Optionally, in an embodiment of the present application, training is performed on the system, and learning is performed by using bayesian personalized ranking, where the specific process is as follows:
constructing a training set triple set for training, wherein the triple set is represented as:
Figure BDA0003060813940000091
wherein
Figure BDA0003060813940000092
For recording the interactions of the user with the bundle package that have occurred in the training set, and
Figure BDA0003060813940000093
is the record of interaction between the user and the bundle that does not appear in the training set, u is the user, i is a bundle clicked by the user u in the training set,? j is a certain bundle package which is not clicked by the user u in the training set;
training by the following function model based on the triplet set:
Figure BDA0003060813940000094
where σ is the Sigmoid function, Θ is the trainable parameter of the model, λ is the coefficient of the $ L2 regular term, σ is the activation function,
Figure BDA0003060813940000095
to the extent user u prefers bundle i,
Figure BDA0003060813940000096
the preference of user u for bundle j.
In order to implement the foregoing embodiments, the present invention further provides a computer device, which 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 bundling recommendation method based on the multi-channel hypergraph neural network according to the foregoing embodiments is implemented.
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.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present application, "plurality" means at least two, e.g., two, three, etc., unless specifically limited otherwise.
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.
The logic and/or steps represented in the flowcharts or otherwise described herein, e.g., an ordered listing of executable instructions that can be considered to implement logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via for instance optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.
It should be understood that portions of the present application may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
It will be understood by those skilled in the art that all or part of the steps carried by the method for implementing the above embodiments may be implemented by hardware related to instructions of a program, which may be stored in a computer readable storage medium, and when the program is executed, the program includes one or a combination of the steps of the method embodiments.
In addition, functional units in the embodiments of the present application may be integrated into one processing module, or each unit may exist alone physically, or two or more units are integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. The integrated module, if implemented in the form of a software functional module and sold or used as a stand-alone product, may also be stored in a computer readable storage medium.
The storage medium mentioned above may be a read-only memory, a magnetic or optical disk, etc. 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 (9)

1. A binding recommendation method based on a multi-channel hypergraph neural network is characterized by comprising the following steps:
acquiring a user-item-bundle package interaction graph and corresponding node characteristics, wherein the user-item-bundle package interaction graph comprises subordinate information of items and bundles, interaction information of users and bundles, and interaction information of users and items;
constructing a user hypergraph, an item hypergraph and a bundle package hypergraph according to the user-item-bundle package interaction graph;
initializing the node characteristics to obtain user characteristics, article characteristics and bundle package characteristics;
carrying out feature extraction on the user features, the article features and the bundle package features and corresponding user hypergraphs, article hypergraphs and bundle package hypergraphs by using hypergraph convolution to obtain new user features, article features and bundle package features;
carrying out feature fusion on the new user features, the article features and the bundle package features to obtain corresponding user feature representation, article feature representation and bundle package feature representation;
aggregating the item feature representations using a bundle-guided item pooling module and adding the aggregated item feature representations to the bundle feature representations to obtain a final representation;
obtaining the preference degree of the user to the bundle package by performing point multiplication on the user characteristic representation and the final representation;
and carrying out binding recommendation according to the preference degree of the binding package.
2. The method of claim 1, wherein the building of the user hypergraph comprises the steps of:
for the interactive information between the user and the articles, taking each article as a center, and connecting the user directly connected with the articles by using the super edges so as to generate a first group of super edges;
for the interaction information of the user and the binding packages, taking each binding package as a center, and connecting the user directly connected with the object by using the super edge so as to generate a second group of super edges;
fusing the first group of the super edges and the second group of the super edges by adopting a splicing fusion mode to generate the user super graph,
the construction of the article hypergraph comprises the following steps:
for the interactive information between the users and the articles, taking each user as a center, and connecting the articles directly connected with the users by using a super edge so as to generate a third group of super edges;
regarding the subordinate information of the user and the binding packages, taking each binding package as a center, and connecting the objects in each binding package by using the super edges so as to generate a fourth group of super edges;
fusing the third group of the super edges and the fourth group of the super edges by adopting a splicing fusion mode to generate the article super graph,
the construction of the bundle hypergraph comprises the following steps:
for the interaction information of the users and the binding packages, taking each user as a center, and connecting the binding packages directly connected with a certain article by using the super edges so as to generate a fifth group of super edges;
for the subordinate information of the articles and the binding packets, connecting the binding packets directly connected with one article by using the super edges so as to generate a sixth group of super edges;
and fusing the fifth group of the super edges and the sixth group of the super edges by adopting a splicing fusion mode to generate the bundle package super graph.
3. The method of claim 1, wherein the user features, item features, bundle features and corresponding user hypergraph, item hypergraph, bundle hypergraph are feature extracted using hypergraph convolution to obtain new user features, item features and bundle features, expressed as:
Figure FDA0003060813930000021
Figure FDA0003060813930000022
Figure FDA0003060813930000023
wherein the content of the first and second substances,
Figure FDA0003060813930000024
and
Figure FDA0003060813930000025
the user characteristic, the item characteristic, and the bundle characteristic, respectively;
Figure FDA0003060813930000026
Figure FDA0003060813930000027
and
Figure FDA0003060813930000028
the corresponding output characteristics;
Figure FDA0003060813930000029
and
Figure FDA00030608139300000210
leakly _ Relu is an activation function for the learnable parameters of the module, HuIn order to allow the user to go beyond the picture,
Figure FDA00030608139300000211
is a degree matrix of the user hypergraph nodes,
Figure FDA00030608139300000212
a matrix, W, which is the superedge of the user hypergraphuWeight matrix for the user hypergraph hyper-edge, HiThe method is characterized in that the method is a super map of an article,
Figure FDA00030608139300000213
is a degree matrix of the nodes of the commodity hypergraph,
Figure FDA00030608139300000214
is a matrix of the over-edges of the hypergraph of the article, WiWeight matrix for the hyper-edge of the item hypergraph, HbIn order to super-map the bundle package,
Figure FDA00030608139300000215
for the degree matrix of the bundle hypergraph nodes,
Figure FDA00030608139300000216
for binding the matrices of the super-edges of the package, WbA weight matrix that is the super edge of the bundle hypergraph.
4. The method of claim 1, wherein the fusing of the characteristics of different levels of users, items, and bundles is represented as:
Figure FDA00030608139300000217
Figure FDA00030608139300000218
Figure FDA00030608139300000219
wherein ijRepresents a user, ikRepresenting an article, bpIndicating a bundle, L being the number of network layers
Figure FDA00030608139300000220
As a feature of the user, it is,
Figure FDA00030608139300000221
is characterized by the characteristics of the article,
Figure FDA00030608139300000222
is a bundle feature.
5. The method of claim 3, wherein the bundle-guided item pooling module is constructed to pool item features within a bundle by:
Figure FDA0003060813930000031
wherein the content of the first and second substances,
Figure FDA0003060813930000032
and
Figure FDA0003060813930000033
to trainable the parameters, σ is an activation function, in particular a LeakyRelu activation function,
Figure FDA0003060813930000034
as a bundle package bpIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure FDA0003060813930000035
is an article ikIs characterized in that it is a mixture of two or more of the above-mentioned components,
Figure FDA0003060813930000036
is an article ilIs characterized by apkFor the weight coefficient of the pooling module, JpSet of items contained in bundle p, ilIs an article il，bpAs a bundle, bpAssembled by articles
Figure FDA0003060813930000037
Is composed of, and article ikIs a set
Figure FDA0003060813930000038
One of the articles;
6. the method of claim 1, wherein the user feature representation and the final representation are dot multiplied by:
Figure FDA0003060813930000039
wherein the content of the first and second substances,
Figure FDA00030608139300000310
for the purpose of the representation of the user characteristic,
Figure FDA00030608139300000311
is the final representation.
7. A binding recommendation system based on a multi-channel hypergraph neural network is characterized by comprising:
the first acquisition module is used for acquiring a user-article-bundle package interaction graph and corresponding node characteristics;
the building module is used for building a user hypergraph, an article hypergraph and a bundle hypergraph according to the user-article-bundle package interaction graph, and initializing according to the node characteristics to obtain user characteristics, article characteristics and bundle package characteristics;
the second acquisition module is used for extracting the user characteristics, the article characteristics and the binding package characteristics and corresponding user hypergraph, article hypergraph and binding package hypergraph by using hypergraph convolution to obtain new user characteristics, article characteristics and binding package characteristics;
the fusion module is used for carrying out feature fusion on the new user features, the article features and the bundle package features to obtain corresponding user feature representation, article feature representation and bundle package feature representation;
a third obtaining module, configured to aggregate the item feature representations using an item pooling module, and add the aggregated item feature representations to the bundle feature representation to obtain a final representation;
and the determining module is used for obtaining the preference degree of the user to the bundle package by performing point multiplication on the user characteristic representation and the final representation.
8. The system of claim 7, wherein the system is trained and learned using bayesian personalized ranking by the specific process of:
constructing a training set triplet set for training, the triplet set represented as:
Figure FDA0003060813930000041
wherein
Figure FDA0003060813930000042
For recording the interactions of the user with the bundle package that have occurred in the training set, and
Figure FDA0003060813930000043
recording the interaction between the user and the bundle package which does not appear in the training set, wherein u is the user, i is a certain bundle package clicked by the user u in the training set, j is a certain bundle package not clicked by the user u in the training set,
training by the following function model based on the triple set:
Figure FDA0003060813930000044
where σ is the Sigmoid function, Θ is the trainable parameter of the model, λ is the coefficient of the $ L2 regular term, σ is the activation function,
Figure FDA0003060813930000045
to the extent user u prefers bundle i,
Figure FDA0003060813930000046
the preference of user u for bundle j.
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 of any one of claims 1-8 when executing the computer program.
CN202110512230.6A 2021-05-11 2021-05-11 Binding recommendation method based on multichannel hypergraph neural network Active CN113177159B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110512230.6A CN113177159B (en) 2021-05-11 2021-05-11 Binding recommendation method based on multichannel hypergraph neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110512230.6A CN113177159B (en) 2021-05-11 2021-05-11 Binding recommendation method based on multichannel hypergraph neural network

Publications (2)

Publication Number Publication Date
CN113177159A true CN113177159A (en) 2021-07-27
CN113177159B CN113177159B (en) 2022-08-05

Family

ID=76929206

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110512230.6A Active CN113177159B (en) 2021-05-11 2021-05-11 Binding recommendation method based on multichannel hypergraph neural network

Country Status (1)

Country Link
CN (1) CN113177159B (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113672811A (en) * 2021-08-24 2021-11-19 广东工业大学 Hypergraph convolution collaborative filtering recommendation method and system based on topology information embedding and computer readable storage medium
CN113705897A (en) * 2021-08-30 2021-11-26 江西鑫铂瑞科技有限公司 Product quality prediction method and system for industrial copper foil production
CN113869992A (en) * 2021-12-03 2021-12-31 平安科技（深圳）有限公司 Artificial intelligence based product recommendation method and device, electronic equipment and medium
CN114329200A (en) * 2021-12-27 2022-04-12 黑龙江大学 Personalized self-adaptive binding recommendation model and recommendation method
CN116204729A (en) * 2022-12-05 2023-06-02 重庆邮电大学 Cross-domain group intelligent recommendation method based on hypergraph neural network
CN116611894A (en) * 2023-06-02 2023-08-18 黑龙江大学 Cross-level graph comparison learning binding commodity recommendation method

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8290818B1 (en) * 2009-11-19 2012-10-16 Amazon Technologies, Inc. System for recommending item bundles
US20140122228A1 (en) * 2012-10-30 2014-05-01 Kelly Joseph Wical Method and system for emergent data processing
US20140358665A1 (en) * 2013-05-31 2014-12-04 Wal-Mart Stores, Inc. Predictive Item Bundling Recommendation Systems And Methods
CN111611472A (en) * 2020-03-31 2020-09-01 清华大学 Binding recommendation method and system based on graph convolution neural network
CN111881350A (en) * 2020-07-23 2020-11-03 清华大学 Recommendation method and system based on mixed graph structured modeling
CN112613602A (en) * 2020-12-25 2021-04-06 神行太保智能科技(苏州)有限公司 Recommendation method and system based on knowledge-aware hypergraph neural network

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8290818B1 (en) * 2009-11-19 2012-10-16 Amazon Technologies, Inc. System for recommending item bundles
US20140122228A1 (en) * 2012-10-30 2014-05-01 Kelly Joseph Wical Method and system for emergent data processing
US20140358665A1 (en) * 2013-05-31 2014-12-04 Wal-Mart Stores, Inc. Predictive Item Bundling Recommendation Systems And Methods
CN111611472A (en) * 2020-03-31 2020-09-01 清华大学 Binding recommendation method and system based on graph convolution neural network
CN111881350A (en) * 2020-07-23 2020-11-03 清华大学 Recommendation method and system based on mixed graph structured modeling
CN112613602A (en) * 2020-12-25 2021-04-06 神行太保智能科技(苏州)有限公司 Recommendation method and system based on knowledge-aware hypergraph neural network

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113672811A (en) * 2021-08-24 2021-11-19 广东工业大学 Hypergraph convolution collaborative filtering recommendation method and system based on topology information embedding and computer readable storage medium
CN113672811B (en) * 2021-08-24 2023-04-18 广东工业大学 Hypergraph convolution collaborative filtering recommendation method and system based on topology information embedding and computer readable storage medium
CN113705897A (en) * 2021-08-30 2021-11-26 江西鑫铂瑞科技有限公司 Product quality prediction method and system for industrial copper foil production
CN113705897B (en) * 2021-08-30 2023-09-26 江西鑫铂瑞科技有限公司 Product quality prediction method and system for industrial copper foil production
CN113869992A (en) * 2021-12-03 2021-12-31 平安科技（深圳）有限公司 Artificial intelligence based product recommendation method and device, electronic equipment and medium
CN114329200A (en) * 2021-12-27 2022-04-12 黑龙江大学 Personalized self-adaptive binding recommendation model and recommendation method
CN114329200B (en) * 2021-12-27 2023-05-26 黑龙江大学 Personalized self-adaptive binding recommendation model and recommendation method
CN116204729A (en) * 2022-12-05 2023-06-02 重庆邮电大学 Cross-domain group intelligent recommendation method based on hypergraph neural network
CN116204729B (en) * 2022-12-05 2024-05-10 武汉光谷康服信息科技有限公司 Cross-domain group intelligent recommendation method based on hypergraph neural network
CN116611894A (en) * 2023-06-02 2023-08-18 黑龙江大学 Cross-level graph comparison learning binding commodity recommendation method
CN116611894B (en) * 2023-06-02 2023-11-17 黑龙江大学 Cross-level graph comparison learning binding commodity recommendation method

Also Published As

Publication number Publication date
CN113177159B (en) 2022-08-05

Similar Documents

Publication Publication Date Title
CN113177159B (en) Binding recommendation method based on multichannel hypergraph neural network
US8983888B2 (en) Efficient modeling system for user recommendation using matrix factorization
CN110717098B (en) Meta-path-based context-aware user modeling method and sequence recommendation method
CN106023015B (en) Course learning path recommendation method and device
CN112487278A (en) Training method of recommendation model, and method and device for predicting selection probability
CN111506820B (en) Recommendation model, recommendation method, recommendation device, recommendation equipment and recommendation storage medium
CN110096617B (en) Video classification method and device, electronic equipment and computer-readable storage medium
CN111046188A (en) User preference degree determining method and device, electronic equipment and readable storage medium
CN110321473B (en) Multi-modal attention-based diversity preference information pushing method, system, medium and device
CN111738780A (en) Method and system for recommending object
CN108427658A (en) A kind of data predication method, device and electronic equipment
CN109218769A (en) A kind of recommended method and relevant device of direct broadcasting room
CN110046301A (en) Object recommendation method and apparatus
CN115358809A (en) Multi-intention recommendation method and device based on graph comparison learning
CN111159279B (en) Model visualization method, device and storage medium
CN114417174A (en) Content recommendation method, device, equipment and computer storage medium
CN113505305A (en) Collaborative filtering recommendation method and system based on decoupling type two-channel hypergraph neural network
CN110990624A (en) Video recommendation method, device, equipment and storage medium
CN110827078A (en) Information recommendation method, device, equipment and storage medium
CN108090110A (en) Recommend software operation to create image and the effect that image is recommended to be acted with demo disk demo software
Mdarbi et al. An approach for selecting cloud service adequate to big data case study: E-health context
CN102577282B (en) Network delay estimation unit and network delay method of estimation
CN113888282A (en) Training method of item recommendation model, item recommendation method and device
CN110264306A (en) Products Show method, apparatus, server and medium based on big data
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
GR01 Patent grant
GR01 Patent grant