CN111611472A - Binding recommendation method and system based on graph convolution neural network - Google Patents

Binding recommendation method and system based on graph convolution neural network Download PDF

Info

Publication number
CN111611472A
CN111611472A CN202010244341.9A CN202010244341A CN111611472A CN 111611472 A CN111611472 A CN 111611472A CN 202010244341 A CN202010244341 A CN 202010244341A CN 111611472 A CN111611472 A CN 111611472A
Authority
CN
China
Prior art keywords
binding
user
item
interaction data
data set
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
CN202010244341.9A
Other languages
Chinese (zh)
Other versions
CN111611472B (en
Inventor
李勇
常健新
高宸
金德鹏
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 CN202010244341.9A priority Critical patent/CN111611472B/en
Publication of CN111611472A publication Critical patent/CN111611472A/en
Application granted granted Critical
Publication of CN111611472B publication Critical patent/CN111611472B/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
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The embodiment of the invention provides a binding recommendation method and system based on a graph convolution neural network. The method comprises the following steps: acquiring historical interaction data of a user and a binding item, historical interaction data of the user and the item and binding and item dependency relationship data; inputting the user interaction probability recommendation result into a binding recommendation model to obtain a user and binding interaction probability recommendation result output by the binding recommendation model; the binding recommendation model is obtained by constructing a unified heterogeneous graph based on a user data set, a binding data set and an article data set, extracting article level graph convolution propagation characteristics and binding level graph convolution propagation characteristics, then performing joint prediction and characteristic connection, and training based on a hard-to-negative sample learning strategy. According to the embodiment of the invention, the interaction relation and the dependency relation among the user, the bundle and the article are reconstructed into the graph, and three associated entity representations are learned from a complex topological structure and high-order connectivity by utilizing the powerful capability of the graph neural network, so that better recommendation performance can be achieved.

Description

Binding recommendation method and system based on graph convolution neural network
Technical Field
The invention relates to the technical field of binding recommendation, in particular to a method and a system for binding recommendation based on a graph convolution neural network.
Background
The binding recommendation is defined as a group of binding articles which are recommended for the user to consume integrally, and the full use of the binding articles on the e-commerce and content platform becomes an important task, so that the user experience can be improved by avoiding the monotonous selection of the user, and the business sales volume can be increased by expanding the order size. Since a bundle is made up of multiple items, the attractiveness of the bundle depends on the items within the bundle, and the attractiveness of each item within the bundle is influenced by the other items displayed together in the bundle. Furthermore, the user needs to be satisfied with most items in the bundle, which means that the interaction between the user and the bundle is more sparse.
While the bundling strategy is currently in widespread use in a variety of locations, little work is done to solve the bundling recommendation problem. Most existing work treats item recommendations and bundle recommendations as two relatively independent tasks and associates them by sharing model parameters or a multi-task framework. The list recommendation model (LIRE) and embedded decomposition machine (EFM) take advantage of user interaction with both merchandise and bundled merchandise simultaneously under the BPR framework. The bundled BPR model (BBPR) utilizes parameters previously learned by the item BPR model. The deep attention multitasking model (DAM) jointly models user-bundle interactions and user-item interactions in a multitasking manner, transferring the benefits of item recommendation tasks into bundle recommendations to alleviate the scarcity of user-bundle interactions.
Currently, several methods are generally adopted:
in a first aspect, a co-selection score is determined for each of a plurality of item pairs, the co-selection score indicating a probability that both of the item pairs are selected by a user, and a co-selection graph is determined between the plurality of items based on the co-selection score. Each node represents an item, and each edge is associated with a co-selection score, thereby transforming the item binding problem into a maximum complete N subgraph optimization problem.
And secondly, calculating the similarity between other items and the current item according to the category or the self attribute of the item which is currently interested by the user, supplementing the most similar item as the current item to serve as a bundle, supplementing other items similar to the items in the bundle one by one to serve as a new bundle, ranking the plurality of item bundles, and selecting one item bundle from the plurality of item bundles to recommend to the user.
It can be easily found that the above scheme has the following limitations: (1) the method of parameter sharing does not explicitly model the relationships between users, items and bundles, and it is difficult to balance the weights between primary and secondary tasks at the loss level in a multitask framework; (2) existing work only considers the correlation between items in a bundle to learn a better item representation to enhance the item recommendation task. However, bundling is a recommendation target, and the similarity between them is more important but neglected; (3) when a user binds with an interaction, his decision psychology has not been taken into account. At the item level, even if the user likes most of the items in the bundle, the user may reject the bundle due to the presence of a disliked item. At the bundle level, for two highly similar bundles, the key to influencing the user's final selection is their non-overlapping parts.
Disclosure of Invention
The embodiment of the invention provides a graph convolution neural network-based binding recommendation method and system, which are used for solving the problems that in the prior art, the weight relation among users, binding and articles cannot be balanced when binding recommendation is carried out, and the corresponding decision reference relation is not comprehensive enough, so that the recommendation result is not accurate and objective enough.
In a first aspect, an embodiment of the present invention provides a graph convolution neural network-based binding recommendation method, including:
acquiring historical interaction data of a user and a binding item, historical interaction data of the user and the item and binding and item dependency relationship data;
inputting the historical interaction data of the user and the binding, the historical interaction data of the user and the item and the dependency relationship data of the binding and the item into a pre-trained binding recommendation model to obtain a recommendation result of the user-binding interaction possibility output by the binding recommendation model; the binding recommendation model is obtained by constructing a unified heterogeneous graph based on a user-binding interaction data set, a binding-item interaction data set and a user-item interaction data set, extracting an item level graph convolution propagation feature and a binding level graph convolution propagation feature, then performing joint prediction and feature connection, and training based on a hard-to-negative sample learning strategy.
Preferably, the bundled recommendation model is obtained by:
acquiring the user and binding interaction data set, the binding and item interaction data set and the user and item interaction data set, and constructing the unified abnormal picture based on the user and binding interaction data set, the binding and item interaction data set and the user and item interaction data set;
extracting the article level map convolution propagation feature and the bundle level map convolution propagation feature based on the uniform abnormal map;
embedding and connecting all layers of the convolution propagation characteristics of the item level map and the convolution propagation characteristics of the binding level map to obtain joint prediction expression of an item propagation visual angle and a binding propagation visual angle;
and training the joint prediction expression by adopting the difficult-to-bear sample learning strategy based on a binding scene to obtain the binding recommendation model.
Preferably, the acquiring the user-bundle interaction data set, the bundle-item interaction data set, and the user-item interaction data set, and constructing the unified heterogeneous composition based on the user-bundle interaction data set, the bundle-item interaction data set, and the user-item interaction data set specifically include:
acquiring a plurality of user information, a plurality of binding information and a plurality of item information, and respectively defining the interaction data of the plurality of user information and the plurality of binding information as the user and binding interaction data set, and the subordinate relationship of the plurality of binding information and the plurality of item information as the binding and item interaction data set and the interaction data of the plurality of user information and the plurality of item information as the user and item interaction data set;
representing the user and bundle interaction data set, the bundle and item interaction data set and the user and item interaction data set by using an undirected graph; wherein the undirected graph comprises nodes and edges, the nodes comprise user nodes, binding nodes and item nodes, and the edges comprise user-binding interaction edges, user-item interaction edges and binding-item dependent edges;
the inputs to the user node, the binding node, and the item node are encoded using one-hot encoding and compressed into dense real-valued vectors.
Preferably, the extracting the article level map convolution propagation feature and the bundle level map convolution propagation feature based on the uniform heteromorphic map specifically includes:
constructing embedding propagation between a user and an article based on the dense real-value vector to obtain an article level embedding updating rule, and obtaining the convolution propagation characteristic of the article level graph according to the article level embedding updating rule;
and constructing embedding propagation between the binding and the user based on the dense real-value vector to obtain a binding level embedding updating rule, and obtaining the convolution propagation characteristic of the binding level graph according to the binding level embedding updating rule.
Preferably, the embedding and connecting all layers of the item level map convolution propagation feature and the bundle level map convolution propagation feature to obtain a joint prediction expression of an item propagation view and a bundle propagation view specifically includes:
carrying out graph convolution propagation on the item level graph convolution propagation characteristics and the binding level graph convolution propagation characteristics for a plurality of times to obtain a plurality of user embedded vectors and a plurality of binding embedded vectors;
and embedding and combining all layers of the user embedded vectors and the binding embedded vectors according to a preset operation mode to obtain the joint prediction expression.
Preferably, the training of the joint prediction expression by using a learning strategy based on a difficult-to-bear sample in a binding scene to obtain the binding recommendation model specifically includes:
defining observed user bundled interaction data and unobserved user bundled interaction data based on the joint predictive expression, constructing paired training data with negative samples based on the observed user bundled interaction data and the unobserved user bundled interaction data;
and taking a preset target function as a model training target, introducing the paired training data according to a preset probability, and training based on the difficult-to-bear sample learning strategy to obtain the binding recommendation model.
Preferably, the training of the bundled recommendation model further comprises setting a number of model hyper-parameters.
In a second aspect, an embodiment of the present invention provides a graph convolution neural network-based binding recommendation system, including:
the acquisition module is used for acquiring historical interaction data of the user and the binding, historical interaction data of the user and the article and the binding and article dependency relationship data;
the processing module is used for inputting the historical interaction data of the user and the binding, the historical interaction data of the user and the article and the dependency relationship data of the binding and the article into a binding recommendation model which is trained in advance to obtain a recommendation result of the interaction possibility between the user and the binding which is output by the binding recommendation model; the binding recommendation model is obtained by constructing a unified heterogeneous graph based on a user-binding interaction data set, a binding-item interaction data set and a user-item interaction data set, extracting an item level graph convolution propagation feature and a binding level graph convolution propagation feature, then performing joint prediction and feature connection, and training based on a hard-to-negative sample learning strategy.
In a third aspect, an embodiment of the present invention provides an electronic device, including:
a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of any one of the graph convolution neural network-based bundling recommendation methods when executing the program.
In a fourth aspect, an embodiment of the present invention provides a non-transitory computer readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps of any one of the graph convolution neural network-based bundling recommendation methods.
According to the binding recommendation method and system based on the graph convolution neural network, provided by the embodiment of the invention, the interaction relation and the dependency relation among the user, the binding and the object are reconstructed into the graph, and the expression of three associated entities is learned from a complex topological structure and high-order connectivity by utilizing the strong capability of the graph neural network, so that better recommendation performance can be achieved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
Fig. 1 is a flowchart of a graph convolution neural network-based bundling recommendation method according to an embodiment of the present invention;
FIG. 2 is a flow chart of a bundle recommendation system according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating a binding recommendation problem definition according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a unified heterogeneous graph structure including users, articles and bundles according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a feature extraction process based on convolution propagation of an item level diagram according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of a process of feature extraction based on bundle level graph convolution propagation according to an embodiment of the present invention;
fig. 7 is a schematic diagram of high-order connectivity in a bundled recommendation scenario according to an embodiment of the present invention;
FIG. 8 is a diagram illustrating a joint prediction process based on two views according to an embodiment of the present invention;
fig. 9 is a schematic diagram of a hard negative sampling process in a bundling scenario according to an embodiment of the present invention;
FIG. 10 is a flowchart illustrating an embodiment of a bundled recommendation system according to the present invention;
fig. 11 is a structural diagram of a binding recommendation system based on a graph-convolution neural network according to an embodiment of the present invention;
fig. 12 is a block diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
To overcome the limitations of the prior art, embodiments of the present invention aim to recommend bundles that a user may interact with based on the user's historical interactions with the item/bundle and the bundle and item composition information. A binding recommendation method based on a graph convolution neural network is provided by considering the binding recommendation problem under scenes such as an e-commerce platform or a content platform, and the interaction and the dependency relationship among users, bindings and commodities are explicitly modeled by unifying the users, the bindings and the commodities into a heterogeneous graph. Collaborative filtering signals between users and items/bundles, semantics of bundles, and similarities between bundles are captured by differentiating dependencies between item nodes and bundle nodes based on two levels of graph convolution propagation over a constructed heterogeneous graph. Given the careful attitude of the user in selecting bundles, the hard-to-negative sampling further encodes a fine-grained representation of the user and bundles.
Fig. 1 is a flowchart of a graph convolution neural network-based bundling recommendation method according to an embodiment of the present invention, as shown in fig. 1, including:
s1, acquiring historical interaction data of the user and the binding, historical interaction data of the user and the article and dependency relationship data of the binding and the article;
s2, inputting the historical interaction data of the user and the binding, the historical interaction data of the user and the article and the dependency relationship data of the binding and the article into a pre-trained binding recommendation model to obtain a recommendation result of the interaction possibility between the user and the binding, which is output by the binding recommendation model; the binding recommendation model is obtained by constructing a unified heterogeneous graph based on a user-binding interaction data set, a binding-item interaction data set and a user-item interaction data set, extracting an item level graph convolution propagation feature and a binding level graph convolution propagation feature, then performing joint prediction and feature connection, and training based on a hard-to-negative sample learning strategy.
Specifically, first, data saved based on a platform is acquired, and historical interaction records of a user on an item and binding configuration information can be acquired, which are embodied as user-binding historical interaction data, user-item historical interaction data and binding-item dependency relationship data, as shown in fig. 2. Here, in order to integrate item level information to improve the accuracy of bundle recommendation, three important pieces of information need to be modeled, using
Figure BDA0002433574690000071
B and I represent a set of users, bundles and items, and define a user-bundle interaction matrix, a user-item interaction matrix and a bundle-item dependency matrix as XM×N＝{xub|u∈U,b∈B}，YM×O＝{yui| U ∈ U, I ∈ I }, and ZN×O＝{zbiI B ∈ B, I ∈ I, each entry is a binary valueubRepresenting the observed interaction y of user u once interacting with bundle buiMeaning that user u has interacted with item i, as shown in fig. 3. Similarly, item zbiMeaning binding b-pack 1Containing item i. M, N and O respectively represent the number of users, the number of bundles and the number of commodities.
Thus, the problem of making a bundle recommendation to a user within a platform can be represented in the form of: user-bundle history interaction XM×NUser-item History interaction YM×OAnd bundle-item dependencies ZN×O(ii) a And (3) outputting: and the binding recommendation model is used for estimating the possibility of the interaction between the user u and the binding b, namely the user-binding interaction possibility recommendation result.
According to the embodiment of the invention, the interaction relation and the dependency relation among the user, the bundle and the article are reconstructed into the graph, and the strong capability of the graph neural network is utilized to learn the representation of three associated entities from the complex topological structure and the high-order connectivity, so that better recommendation performance can be achieved.
Based on the above embodiment, the bundled recommendation model is obtained by the following steps:
acquiring the user and binding interaction data set, the binding and item interaction data set and the user and item interaction data set, and constructing the unified abnormal picture based on the user and binding interaction data set, the binding and item interaction data set and the user and item interaction data set;
extracting the article level map convolution propagation feature and the bundle level map convolution propagation feature based on the uniform abnormal map;
embedding and connecting all layers of the convolution propagation characteristics of the item level map and the convolution propagation characteristics of the binding level map to obtain joint prediction expression of an item propagation visual angle and a binding propagation visual angle;
and training the joint prediction expression by adopting the difficult-to-bear sample learning strategy based on a binding scene to obtain the binding recommendation model.
Specifically, a binding recommendation model is established, namely, a user is modeled uniformly, a uniform abnormal graph is constructed according to the complex relation of the object and the binding, graph convolution propagation is carried out on the basis of the object level to extract features, graph convolution propagation is carried out on the basis of the binding level to extract features, prediction is carried out on the basis of two propagation visual angles of the object and the binding, and finally model training which is difficult to carry out negative sampling under a binding scene is carried out.
Based on any of the above embodiments, the obtaining the user-bundle interaction data set, the bundle-item interaction data set, and the user-item interaction data set, and constructing the unified heterogeneous map based on the user-bundle interaction data set, the bundle-item interaction data set, and the user-item interaction data set specifically include:
acquiring a plurality of user information, a plurality of binding information and a plurality of item information, and respectively defining the interaction data of the plurality of user information and the plurality of binding information as the user and binding interaction data set, and the subordinate relationship of the plurality of binding information and the plurality of item information as the binding and item interaction data set and the interaction data of the plurality of user information and the plurality of item information as the user and item interaction data set;
representing the user and bundle interaction data set, the bundle and item interaction data set and the user and item interaction data set by using an undirected graph; wherein the undirected graph comprises nodes and edges, the nodes comprise user nodes, binding nodes and item nodes, and the edges comprise user-binding interaction edges, user-item interaction edges and binding-item dependent edges;
the inputs to the user node, the binding node, and the item node are encoded using one-hot encoding and compressed into dense real-valued vectors.
Specifically, to explicitly model the relationship between the user, the bundle and the item, a unified heterogeneous graph is first constructed. The interaction and dependency data may be represented by an undirected graph G ═ V, E, where the nodes are
Figure BDA0002433574690000095
By user nodes
Figure BDA0002433574690000094
Binding node B ∈ B and item node I ∈ I, edge EIs formed by corresponding to xubUser-binding interaction edge (u, b) of 1, corresponding to yuiUser-item interaction edge (u, i) and corresponding z ═ 1biBundle-item dependent edges (b, i) of 1. The heterogeneous graph construction process for uniformly modeling user, item and bundle complex relationships is shown in FIG. 4.
For users, items and binding nodes on the construction graph, the input is further encoded with one-hot encoding and then compressed into a dense real-valued vector, as follows:
Figure BDA0002433574690000091
wherein
Figure BDA0002433574690000092
Representing the one-hot feature vectors of user u, item i and bundle b. P, Q and R represent matrices for user embedding, item embedding and binding embedding, respectively.
Based on any of the above embodiments, the extracting, based on the unified heterogeneous map, the item level map convolution propagation feature and the bundle level map convolution propagation feature specifically includes:
constructing embedding propagation between a user and an article based on the dense real-value vector to obtain an article level embedding updating rule, and obtaining the convolution propagation characteristic of the article level graph according to the article level embedding updating rule;
and constructing embedding propagation between the binding and the user based on the dense real-value vector to obtain a binding level embedding updating rule, and obtaining the convolution propagation characteristic of the binding level graph according to the binding level embedding updating rule.
In particular, in one aspect, a user's preference for items in a bundle can attract the user's attention and interest in the bundle. Because bundled items are typically carefully designed, they are generally functionally compatible with each other and constitute some semantics to influence the user's selection context. For example, a binding with a mattress and a bed frame reflects the meaning of a bedroom home, while a binding with a suit and a tie reflects the meaning of a workplace dress.
To capture the user's preferences for an item and the characteristics of the item itself, an embedded propagation between the user and the item is first constructed. Pooling of information from item to bundle may then capture semantic information of the bundle from the item hierarchy. The propagation and pooling based embedded update rules for user u, item i and bundle b can be expressed as:
Figure BDA0002433574690000093
Figure BDA0002433574690000101
Figure BDA0002433574690000102
Figure BDA0002433574690000103
wherein W1Is a learnable weight, b1Is a learnable offset, σ is a nonlinear activation function, LeakyReLU.
Figure BDA0002433574690000104
Representing the neighbors of user u, item i, and bundle b, respectively. The aggregation function aggregate may be a function such as a simple mean function, a mean function with sampling, maximum pooling, and the like. Through the special propagation mechanism, the influence of the sparsity of the binding data can be reduced, and the cold start capability of model processing is naturally improved. The process of feature extraction based on article hierarchy map convolution propagation is shown in fig. 5.
On the other hand, the close association between items in a bundle makes two bundles sharing some items very similar. The degree of similarity can be inferred from how many items they share. For example, a computer installed suit sharing five components is closer in performance than two, and a movie list sharing ten movies is closer in theme than five. Bundles that share more items may often be considered simultaneously for the user.
Graph convolution embedding propagation bound to the user is first performed to learn the user's preferences at the binding level. Then, user-to-bundle embedded propagation is performed to extract the overall properties of the bundle. Since highly overlapping bundles exhibit a similar pattern in attracting users, a weighted embedding propagation is performed on the bundle-item-bundle path, with the degree of overlap of the bundles as a weight, to capture the alternative relationships between the bundles. The binding-level embedded update rule may be expressed as:
Figure BDA0002433574690000105
Figure BDA0002433574690000106
Figure BDA0002433574690000107
wherein W2And b2Respectively, a trainable transformation matrix and an offset. Mlβ representing the neighbors of bundle b on the bundle-item-bundle pathbb′Indicating the degree of overlap between the normalized bundles. Propagation between the properties represented by similar b helps bundle learning to better represent and further enhances messaging between u and b. The process of feature extraction based on bundle level graph convolution propagation is shown in fig. 6.
Based on any of the above embodiments, the embedding and connecting all layers of the article-level graph convolution propagation feature and the bundle-level graph convolution propagation feature to obtain a joint prediction expression of an article propagation view angle and a bundle propagation view angle specifically includes:
carrying out graph convolution propagation on the item level graph convolution propagation characteristics and the binding level graph convolution propagation characteristics for a plurality of times to obtain a plurality of user embedded vectors and a plurality of binding embedded vectors;
and embedding and combining all layers of the user embedded vectors and the binding embedded vectors according to a preset operation mode to obtain the joint prediction expression.
In particular, since it is desirable to utilize graph neural networks to learn high-order connectivity of user-bundle interactions, user-item interactions and item-bundle dependencies. Particularly higher order connectivity of dependencies, which has never been considered before. E.g. binding b1And binding b2Shared article i1Bundle b2And binding b3Shared article i2Despite the binding b1And binding b3Do not share any items, but they are somewhat similar, as shown in fig. 7. Specifically, L times of graph convolution propagation are iteratively performed, obtaining L user/bundled embedded vectors. And all layers are embedded and combined, and a preset operation mode, such as a connection or summation mode, is adopted to combine information received from neighbors at different depths for prediction. The prediction process based on both propagation views of the item and the bundle is shown in fig. 8.
Figure BDA0002433574690000111
Figure BDA0002433574690000112
Figure BDA0002433574690000113
Figure BDA0002433574690000114
Because the special design of hierarchical propagation can distinguish not only the interaction or the dependency relationship between any two nodes in the unified heterogeneous graph, but also whether the item node i belongs to the binding node b or the binding node b belongs to the item node i, the information of two hierarchies needs to be considered simultaneously during prediction. Specifically, the final prediction is made by user and binding embedding, such as a simple inner product approach, and combining the views of the two levels of items and bindings, such as a simple summation approach, for example as follows:
Figure BDA0002433574690000115
based on any one of the above embodiments, the training of the joint prediction expression by using a learning strategy based on a difficult-to-bear sample in a binding scenario to obtain the binding recommendation model specifically includes:
defining observed user bundled interaction data and unobserved user bundled interaction data based on the joint predictive expression, constructing paired training data with negative samples based on the observed user bundled interaction data and the unobserved user bundled interaction data;
and taking a preset target function as a model training target, introducing the paired training data according to a preset probability, and training based on the difficult-to-bear sample learning strategy to obtain the binding recommendation model.
In particular, because the bundle contains more items and has a higher price, the user is often very careful when making decisions or spending money in the bundling scenario to avoid unnecessary risk. For example, even if the user likes most of the items in the bundle, the bundle may be rejected because there is a disliked item. Furthermore, for two highly similar bundles, the key to influencing the user's final selection is their non-overlapping parts.
In order to optimize the model, the decision of the user in the binding interaction is considered, and a learning strategy based on the hard negative sample in the binding scene is designed. Firstly, a pair-wise learning mode is adopted, and the mode is widely applied to an implicit recommendation system. Then after the model converges, a difficult negative sample is introduced with a certain probability for more elaborate training. Thus, the objective function is defined as follows:
Figure BDA0002433574690000121
wherein
Figure BDA0002433574690000122
Representing paired training data with negative samples.
Figure BDA0002433574690000123
And
Figure BDA0002433574690000124
representing observed and unobserved user binding interactions, respectively. In the hard negative sampler, for each
Figure BDA0002433574690000125
Is a bundle with which u does not interact but interacts with most of the items inside or overlaps with b. λ is the weight of the L2 regularization term and Θ is the trainable set of parameters. The hard negative sample is constructed in the manner shown in fig. 9.
Based on any of the above embodiments, the training of the bundled recommendation model further comprises setting a number of model hyper-parameters.
Specifically, when the bundled recommendation model is trained, model hyper-parameters also need to be set, including a negative sample number sample _ number, a batch size mini _ batch _ size, an embedding size embedding _ size, a learning rate learning _ rate, an L2 regular item L2_ normalization, a message loss rate message _ drop and a node loss rate node _ drop, and a selection probability hard _ rate of a hard negative sample. In the process of training the network, the weights and bias values of each layer of the network can be updated by a Stochastic gradient descent method (Stochastic gradientdecision) in the process of back propagation.
In order to more clearly illustrate the embodiments of the present invention, a specific implementation of the embodiments of the present invention will be described below with reference to fig. 10.
The first implementation mode comprises the following steps: as shown in the middle branch of FIG. 10, the user wants to recommend a new bundle to the user using the platform-tracked user interaction with the item and bundle history. The platform can be any e-commerce and content platform, corresponding to any article capable of forming a bundle, such as goods, food, places, music, books, movies, news, etc.
Firstly, historical interaction of a user and an article, historical interaction of the user and a binding and binding constituting information are formalized into a matrix, and a user-binding historical interaction matrix X is obtainedM×NUser-item history interaction matrix YM×OAnd bundle-item dependency matrix ZN×OA uniform heteromorphic graph can be described by three matrices. Wherein the node
Figure BDA00024335746900001314
By user nodes
Figure BDA00024335746900001315
Binding node B ∈ B and item node I ∈ I, edge E is composed of the corresponding xubUser-binding interaction edge (u, b) of 1, corresponding to yuiUser-item interaction edge (u, i) and corresponding z ═ 1biBundle-item dependent edges (b, i) of 1. For users, items and binding nodes on the construction graph, the input is encoded with one-hot encoding and then compressed into a dense real-valued vector:
Figure BDA0002433574690000131
wherein
Figure BDA0002433574690000132
Representing the one-hot feature vectors of user u, item i and bundle b. P, Q, and R represent matrices of learnable user embedding, item embedding, and bundle embedding, respectively. The input may be one-hot coded here with only two types of interactive and dependent data, and the coded representation may be enhanced with these additional features when other properties of the user, item and bundle are available in the platform (e.g. user image such as age, gender, etc., item/bundle properties such as price, name, picture, etc.).
The input features of the user, the article and the bundle are represented as layer 0 features of a graph neural network, graph structure information is captured by graph convolution propagation on a graph structure, and entity features are updated from the aspect of representation learning. For the propagation of the item hierarchy, the user node u of the l-th layer is firstly propagatedTo represent
Figure BDA0002433574690000133
And representation of its article neighbor node i
Figure BDA0002433574690000134
Aggregating to user node u to obtain a representation of user node u at layer l +1
Figure BDA0002433574690000135
Then the item node i of the l-th layer is represented
Figure BDA0002433574690000136
And representation of its user neighbor node u
Figure BDA0002433574690000137
Aggregate to item node i to obtain a representation of item node i at level l +1
Figure BDA0002433574690000138
Then, the representation of the l +1 st tier binding node b
Figure BDA0002433574690000139
Then the node i is neighbored by its article representation
Figure BDA00024335746900001310
Polymerized to obtain the polymer. The aggregation function is not limited to functions such as simple mean function, mean function with sampling, maximum pooling, etc. The process of feature extraction based on article hierarchy map convolution propagation is shown in fig. 5. For the propagation of the binding hierarchy, the representation of the user node u at the l-th level is first represented
Figure BDA00024335746900001311
And its representation of binding neighbor node b
Figure BDA00024335746900001312
Aggregating to user node u to obtain a representation of user node u at layer l +1
Figure BDA00024335746900001313
Then the representation of the binding node b of the l layer is carried out
Figure BDA0002433574690000141
Representation of user neighbor node u
Figure BDA0002433574690000142
And representation of binding neighbor node b' on the binding-article-binding element path
Figure BDA0002433574690000143
Aggregating to bundled nodes b to obtain a representation of bundled nodes b for layer l +1
Figure BDA0002433574690000144
Wherein aggregation of the binder path neighbors based on overlap takes the degree of overlap between the bindings as a weight. The process of feature extraction based on bundle level graph convolution propagation is shown in fig. 6.
After L times of graph convolution propagation are carried out iteratively, L embedding vectors of users/bundles are obtained, embedding of all layers is combined, including but not limited to a connection mode or a summation mode, and final representations of the users and the bundles under two propagation view angles are obtained
Figure BDA0002433574690000145
Then, the final prediction is made by embedding the user and the bundle, including but not limited to inner product mode, and combining the view angles of the two layers of the article and the bundle, including but not limited to summation mode, to obtain the possibility that the user u interacts with the bundle b
Figure BDA0002433574690000146
The prediction process based on both propagation views of the item and the bundle is shown in fig. 8. For model training, a pair-wise learning mode widely applied to an implicit recommendation system is adopted to enable the estimated probability score of the observed user bundle interaction to be larger than the probability score of the unobserved user bundle interaction. Then after the model convergence, with a certain ruleThe rate (e.g., 80%) introduces more elaborate training based on the difficult-to-negative examples in the bundling scenario, selecting bundles in each positive sample pair that do not interact with the user but interact with most of their internal items as difficult-to-negative examples, or selecting other bundles in each positive sample pair that have overlapping items with the bundle as difficult-to-negative examples. The estimated likelihood score of an observed user binding interaction is made greater than the likelihood score of a user binding sample pair that was not observed but was difficult for the user to decide. And optimizing the objective function by a random gradient descent method to obtain all learnable parameters in the model, thereby obtaining an end-to-end binding recommendation system.
The second embodiment: as shown in the left branch of FIG. 10, the user wants to recommend a new bundle to the user using the platform-tracked user's interactions with the bundle history. The platform can be any e-commerce and content platform, corresponding to any article capable of forming a bundle, such as goods, food, places, music, books, movies, news, etc.
Firstly, historical interaction of users and binding construction information are formalized into a matrix, and a user-binding historical interaction matrix X is obtainedM×NAnd bundle-item dependency matrix ZN×OA uniform heteromorphic graph can be described by both matrices. Wherein the node
Figure BDA0002433574690000148
By user nodes
Figure BDA0002433574690000147
Binding node B ∈ B and item node I ∈ I, edge E is composed of the corresponding xubUser-bundle interaction edge (u, b) and corresponding z for 1biBundle-item dependent edges (b, i) of 1. For users and bundled nodes on the construction graph, the input is encoded using one-hot encoding and then compressed into dense real-valued vectors:
Figure BDA0002433574690000151
Figure BDA0002433574690000152
wherein
Figure BDA0002433574690000153
Representing the one-hot feature vectors of user u and bundle b. P and R represent matrices of learnable user embedding and binding embedding, respectively. The input can be encoded one-hot here only through the interaction and dependency data, and when the user and other properties of the bundle are available in the platform (e.g., user image such as age, gender, etc., bundle properties such as price, name, picture, etc.), the encoded representation can be enhanced with these additional features.
Representing the user and the bundled input features as layer 0 features of a graph neural network, acquiring graph structure information by performing graph convolution propagation on a graph structure, and updating entity features from the aspect of representation learning. For the propagation of the binding hierarchy, the representation of the user node u at level 1 is first represented
Figure BDA0002433574690000154
And its representation of binding neighbor node b
Figure BDA0002433574690000155
Aggregating to user node u to obtain a representation of user node u at layer l +1
Figure BDA0002433574690000156
Then the representation of the binding node b of the 1 st layer
Figure BDA0002433574690000157
Representation of user neighbor node u
Figure BDA0002433574690000158
And representation of binding neighbor node b' on the binding-article-binding element path
Figure BDA0002433574690000159
Aggregating to bundled nodes b to obtain a representation of bundled nodes b for layer l +1
Figure BDA00024335746900001510
Wherein aggregation of the binder path neighbors based on overlap takes the degree of overlap between the bindings as a weight. The process of feature extraction based on bundle level graph convolution propagation is shown in fig. 6.
After iteratively performing L times of graph convolution propagation, L user/bundled embedded vectors are obtained. And combines the embedding of all layers, including but not limited to concatenation or summation, to obtain a final representation of the user and bundle from a propagation perspective
Figure BDA00024335746900001511
Then, a final prediction is made through embedding of the user and the bundle, including but not limited to inner product, to obtain the possibility that the user u interacts with the bundle b
Figure BDA00024335746900001512
For model training, a pair-wise learning mode widely applied to an implicit recommendation system is adopted to enable the estimated probability score of the observed user bundle interaction to be larger than the probability score of the unobserved user bundle interaction. Then after the model converges, a certain probability (such as 80%) is introduced to perform more elaborate training based on the hard negative sample in the binding scene, and for the binding in each positive sample pair, the other binding with the overlapping article with the binding is selected as the hard negative sample. The estimated likelihood score of an observed user binding interaction is made greater than the likelihood score of a user binding sample pair that was not observed but was difficult for the user to decide. And optimizing the objective function by a random gradient descent method to obtain all learnable parameters in the model, thereby obtaining an end-to-end binding recommendation system.
The third embodiment is as follows: as shown in the right branch of FIG. 10, the user wants to recommend a new bundle to the user using the platform-tracked user's interactions with the item history. The platform can be any e-commerce and content platform, corresponding to any article capable of forming a bundle, such as goods, food, places, music, books, movies, news, etc.
Firstly, the methodThe historical interaction and binding information of the user and the article is formalized into a matrix, and a user-article historical interaction matrix Y is obtainedM×OAnd bundle-item dependency matrix ZN×OA uniform heteromorphic graph can be described by both matrices. Wherein the node
Figure BDA00024335746900001615
By user nodes
Figure BDA0002433574690000161
Binding node B ∈ B and item node I ∈ I, edge E is formed by the correspondence of yuiUser-item interaction edge (u, i) and corresponding z ═ 1biBundle-item dependent edges (b, i) of 1. For user and item nodes on the construction graph, the input is encoded using one-hot encoding and then compressed into dense real-valued vectors:
Figure BDA0002433574690000162
Figure BDA0002433574690000163
wherein
Figure BDA0002433574690000164
Representing the one-hot feature vector of user u, item i. P, Q represent matrices for learnable user embedding and item embedding, respectively. The input may be encoded one-hot only by the interaction and dependency data, and the encoded representation may be enhanced with these additional features when other attributes of the user and item are available in the platform (e.g., user image such as age, gender, etc., item attributes such as price, name, picture, etc.).
The input features of the user and the article are represented as layer 0 features of a graph neural network, graph structure information is captured by graph convolution propagation on a graph structure, and entity features are updated from the aspect of representation learning. For the propagation of the item hierarchy, the representation of the user node u at the l-th level is first represented
Figure BDA0002433574690000165
And representation of its article neighbor node i
Figure BDA0002433574690000166
Aggregating to user node u to obtain a representation of user node u at layer l +1
Figure BDA0002433574690000167
Then the item node i of the l-th layer is represented
Figure BDA0002433574690000168
And representation of its user neighbor node u
Figure BDA0002433574690000169
Aggregate to item node i to obtain a representation of item node i at level l +1
Figure BDA00024335746900001610
Then, the representation of the l +1 st tier binding node b
Figure BDA00024335746900001611
Then the node i is neighbored by its article representation
Figure BDA00024335746900001612
Polymerized to obtain the polymer. The aggregation function is not limited to functions such as simple mean function, mean function with sampling, maximum pooling, etc. The process of feature extraction based on article hierarchy map convolution propagation is shown in fig. 5.
After iteratively performing L times of graph convolution propagation, L user/bundled embedded vectors are obtained. And combines the embedding of all layers, including but not limited to concatenation or summation, to obtain a final representation of the user and bundle from a propagation perspective
Figure BDA00024335746900001613
Then, a final prediction is made through embedding of the user and the bundle, including but not limited to inner product, to obtain the possibility that the user u interacts with the bundle b
Figure BDA00024335746900001614
For model training, a pair-wise learning mode widely applied to an implicit recommendation system is adopted to enable the estimated probability score of the observed user bundle interaction to be larger than the probability score of the unobserved user bundle interaction. Then after the model converges, a certain probability (such as 80%) is introduced to perform more detailed training based on the hard negative sample in the binding scenario, and for the users in each positive sample pair, the binding which does not interact with the user but interacts with most of the objects in the user is selected as the hard negative sample. The estimated likelihood score of an observed user binding interaction is made greater than the likelihood score of a user binding sample pair that was not observed but was difficult for the user to decide. And optimizing the objective function by a random gradient descent method to obtain all learnable parameters in the model, thereby obtaining an end-to-end binding recommendation system.
Fig. 11 is a structural diagram of a binding recommendation system based on a graph convolution neural network according to an embodiment of the present invention, as shown in fig. 11, including: an acquisition module 1101 and a processing module 1102; wherein:
the obtaining module 1101 is configured to obtain user-binding historical interaction data, user-item historical interaction data, and binding-item dependency relationship data; the processing module 1102 is configured to input the historical user-to-bundle interaction data, the historical user-to-article interaction data, and the binding-to-article dependency relationship data into a pre-trained binding recommendation model, so as to obtain a recommendation result of user-to-binding interaction possibility output by the binding recommendation model; the binding recommendation model is obtained by constructing a unified heterogeneous graph based on a user-binding interaction data set, a binding-item interaction data set and a user-item interaction data set, extracting an item level graph convolution propagation feature and a binding level graph convolution propagation feature, then performing joint prediction and feature connection, and training based on a hard-to-negative sample learning strategy.
The system provided by the embodiment of the present invention is used for executing the corresponding method, the specific implementation manner of the system is consistent with the implementation manner of the method, and the related algorithm flow is the same as the algorithm flow of the corresponding method, which is not described herein again.
According to the embodiment of the invention, the interaction relation and the dependency relation among the user, the bundle and the article are reconstructed into the graph, and the strong capability of the graph neural network is utilized to learn the representation of three associated entities from the complex topological structure and the high-order connectivity, so that better recommendation performance can be achieved.
Fig. 12 illustrates a physical structure diagram of an electronic device, which may include, as shown in fig. 12: a processor (processor)1210, a communication Interface (Communications Interface)1220, a memory (memory)1230, and a communication bus 1240, wherein the processor 1210, the communication Interface 1220, and the memory 1230 communicate with each other via the communication bus 1240. Processor 1210 may call logic instructions in memory 1230 to perform the following method: acquiring historical interaction data of a user and a binding item, historical interaction data of the user and the item and binding and item dependency relationship data; inputting the historical interaction data of the user and the binding, the historical interaction data of the user and the item and the binding-item dependency relationship data into a pre-trained binding recommendation model to obtain a recommendation result of the interaction possibility between the user and the binding, which is output by the binding recommendation model; the binding recommendation model is obtained by constructing a unified heterogeneous graph based on a user-binding interaction data set, a binding-item interaction data set and a user-item interaction data set, extracting an item level graph convolution propagation feature and a binding level graph convolution propagation feature, then performing joint prediction and feature connection, and training based on a hard-to-negative sample learning strategy.
In addition, the logic instructions in the memory 1230 may be implemented in software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as a stand-alone product. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In another aspect, an embodiment of the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, where the computer program is implemented to perform the transmission method provided in the foregoing embodiments when executed by a processor, and for example, the method includes: acquiring historical interaction data of a user and a binding item, historical interaction data of the user and the item and binding and item dependency relationship data; inputting the historical interaction data of the user and the binding, the historical interaction data of the user and the item and the dependency relationship data of the binding and the item into a pre-trained binding recommendation model to obtain a recommendation result of the interaction possibility between the user and the binding, which is output by the binding recommendation model; the binding recommendation model is obtained by constructing a unified heterogeneous graph based on a user-binding interaction data set, a binding-item interaction data set and a user-item interaction data set, extracting an item level graph convolution propagation feature and a binding level graph convolution propagation feature, then performing joint prediction and feature connection, and training based on a hard-to-negative sample learning strategy.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (10)

1. A binding recommendation method based on a graph convolution neural network is characterized by comprising the following steps:
acquiring historical interaction data of a user and a binding item, historical interaction data of the user and the item and binding and item dependency relationship data;
inputting the historical interaction data of the user and the binding, the historical interaction data of the user and the item and the dependency relationship data of the binding and the item into a pre-trained binding recommendation model to obtain a recommendation result of the interaction possibility between the user and the binding, which is output by the binding recommendation model; the binding recommendation model is obtained by constructing a unified heterogeneous graph based on a user-binding interaction data set, a binding-item interaction data set and a user-item interaction data set, extracting an item level graph convolution propagation feature and a binding level graph convolution propagation feature, then performing joint prediction and feature connection, and training based on a hard-to-negative sample learning strategy.
2. The graph convolution neural network-based bundled recommendation method according to claim 1, wherein the bundled recommendation model is obtained by:
acquiring the user and binding interaction data set, the binding and item interaction data set and the user and item interaction data set, and constructing the unified abnormal picture based on the user and binding interaction data set, the binding and item interaction data set and the user and item interaction data set;
extracting the article level map convolution propagation feature and the bundle level map convolution propagation feature based on the uniform abnormal map;
embedding and connecting all layers of the convolution propagation characteristics of the item level map and the convolution propagation characteristics of the binding level map to obtain joint prediction expression of an item propagation visual angle and a binding propagation visual angle;
and training the joint prediction expression by adopting the difficult-to-bear sample learning strategy based on a binding scene to obtain the binding recommendation model.
3. The graph convolutional neural network-based binding recommendation method of claim 2, wherein the obtaining the user-to-binding interaction data set, the binding-to-item interaction data set, and the user-to-item interaction data set, and the constructing the uniform heteromorphic graph based on the user-to-binding interaction data set, the binding-to-item interaction data set, and the user-to-item interaction data set specifically comprise:
acquiring a plurality of user information, a plurality of binding information and a plurality of item information, and respectively defining the interaction data of the plurality of user information and the plurality of binding information as the user and binding interaction data set, and the subordinate relationship of the plurality of binding information and the plurality of item information as the binding and item interaction data set and the interaction data of the plurality of user information and the plurality of item information as the user and item interaction data set;
representing the user and bundle interaction data set, the bundle and item interaction data set and the user and item interaction data set by using an undirected graph; wherein the undirected graph comprises nodes and edges, the nodes comprise user nodes, binding nodes and item nodes, and the edges comprise user-binding interaction edges, user-item interaction edges and binding-item dependent edges;
the inputs to the user node, the binding node, and the item node are encoded using one-hot encoding and compressed into dense real-valued vectors.
4. The graph convolution neural network-based binding recommendation method according to claim 3, wherein the extracting the item level graph convolution propagation feature and the binding level graph convolution propagation feature based on the uniform anomaly graph specifically includes:
constructing embedding propagation between a user and an article based on the dense real-value vector to obtain an article level embedding updating rule, and obtaining the convolution propagation characteristic of the article level graph according to the article level embedding updating rule;
and constructing embedding propagation between the binding and the user based on the dense real-value vector to obtain a binding level embedding updating rule, and obtaining the convolution propagation characteristic of the binding level graph according to the binding level embedding updating rule.
5. The graph convolution neural network-based binding recommendation method according to claim 2, wherein the embedding connection of all layers is performed on the item-level graph convolution propagation feature and the binding-level graph convolution propagation feature to obtain a joint prediction expression of an item propagation perspective and a binding propagation perspective, and specifically includes:
carrying out graph convolution propagation on the item level graph convolution propagation characteristics and the binding level graph convolution propagation characteristics for a plurality of times to obtain a plurality of user embedded vectors and a plurality of binding embedded vectors;
and embedding and combining all layers of the user embedded vectors and the binding embedded vectors according to a preset operation mode to obtain the joint prediction expression.
6. The graph convolution neural network-based binding recommendation method according to claim 2, wherein the training of the joint prediction expression by using a difficult-to-negative sample learning strategy based on a binding scenario to obtain the binding recommendation model specifically comprises:
defining observed user bundled interaction data and unobserved user bundled interaction data based on the joint predictive expression, constructing paired training data with negative samples based on the observed user bundled interaction data and the unobserved user bundled interaction data;
and taking a preset target function as a model training target, introducing the paired training data according to a preset probability, and training based on the difficult-to-bear sample learning strategy to obtain the binding recommendation model.
7. The graph-convolution neural network-based bundled recommendation method of any one of claims 1-6, wherein the training of the bundled recommendation model further includes setting a number of model hyper-parameters.
8. A graph convolution neural network-based binding recommendation system is characterized by comprising:
the acquisition module is used for acquiring historical interaction data of the user and the binding, historical interaction data of the user and the article and the binding and article dependency relationship data;
the processing module is used for inputting the historical interaction data of the user and the binding, the historical interaction data of the user and the article and the dependency relationship data of the binding and the article into a binding recommendation model which is trained in advance to obtain a recommendation result of the interaction possibility between the user and the binding which is output by the binding recommendation model; the binding recommendation model is obtained by constructing a unified heterogeneous graph based on a user-binding interaction data set, a binding-item interaction data set and a user-item interaction data set, extracting an item level graph convolution propagation feature and a binding level graph convolution propagation feature, then performing joint prediction and feature connection, and training based on a hard-to-negative sample learning strategy.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program performs the steps of the graph convolution based neural network binding recommendation method of any one of claims 1 to 7.
10. A non-transitory computer readable storage medium, having stored thereon a computer program, wherein the computer program, when being executed by a processor, implements the steps of the graph convolution neural network-based bundling recommendation method according to any one of claims 1 to 7.
CN202010244341.9A 2020-03-31 2020-03-31 Binding recommendation method and system based on graph convolution neural network Active CN111611472B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010244341.9A CN111611472B (en) 2020-03-31 2020-03-31 Binding recommendation method and system based on graph convolution neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010244341.9A CN111611472B (en) 2020-03-31 2020-03-31 Binding recommendation method and system based on graph convolution neural network

Publications (2)

Publication Number Publication Date
CN111611472A true CN111611472A (en) 2020-09-01
CN111611472B CN111611472B (en) 2024-02-27

Family

ID=72200320

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010244341.9A Active CN111611472B (en) 2020-03-31 2020-03-31 Binding recommendation method and system based on graph convolution neural network

Country Status (1)

Country Link
CN (1) CN111611472B (en)

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112085171A (en) * 2020-09-15 2020-12-15 中山大学 Recommendation method based on clustering multi-entity graph neural network
CN112084407A (en) * 2020-09-08 2020-12-15 辽宁工程技术大学 Collaborative filtering recommendation method fusing graph neural network and attention mechanism
CN112465593A (en) * 2020-11-27 2021-03-09 中国科学技术大学 Method for realizing fashion suit recommendation through graph neural network
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
CN112651487A (en) * 2020-12-21 2021-04-13 广东交通职业技术学院 Data recommendation method, system and medium based on graph collapse convolution neural network
CN112650944A (en) * 2020-12-29 2021-04-13 哈尔滨理工大学 Collaborative filtering recommendation algorithm based on high-order connection graph
CN112686147A (en) * 2020-12-30 2021-04-20 北京纵目安驰智能科技有限公司 Vehicle and wheel dependency prediction method and system, storage medium and terminal
CN112765373A (en) * 2021-01-29 2021-05-07 北京达佳互联信息技术有限公司 Resource recommendation method and device, electronic equipment and storage medium
CN112925977A (en) * 2021-02-26 2021-06-08 中国科学技术大学 Recommendation method based on self-supervision graph representation learning
CN113065045A (en) * 2021-04-20 2021-07-02 支付宝(杭州)信息技术有限公司 Method and device for carrying out crowd division and training multitask model on user
CN113177159A (en) * 2021-05-11 2021-07-27 清华大学 Binding recommendation method based on multichannel hypergraph neural network
CN113343121A (en) * 2021-06-02 2021-09-03 合肥工业大学 Lightweight graph convolution collaborative filtering recommendation method based on multi-granularity popularity characteristics
CN113468227A (en) * 2021-06-25 2021-10-01 北京达佳互联信息技术有限公司 Information recommendation method, system, device and storage medium based on graph neural network
CN113763014A (en) * 2021-01-05 2021-12-07 北京沃东天骏信息技术有限公司 Article co-occurrence relation determining method and device and judgment model obtaining method and device
CN114065048A (en) * 2021-11-25 2022-02-18 西安电子科技大学 Article recommendation method based on multi-different-pattern neural network
CN114329200A (en) * 2021-12-27 2022-04-12 黑龙江大学 Personalized self-adaptive binding recommendation model and recommendation method
CN114399028A (en) * 2022-01-14 2022-04-26 马上消费金融股份有限公司 Information processing method, graph convolution neural network training method and electronic equipment
CN116611894A (en) * 2023-06-02 2023-08-18 黑龙江大学 Cross-level graph comparison learning binding commodity recommendation method

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108713212A (en) * 2016-03-04 2018-10-26 范德曼公司 Method, system and medium for providing information based on grouping information
CN108874914A (en) * 2018-05-29 2018-11-23 吉林大学 A kind of information recommendation method based on the long-pending and neural collaborative filtering of picture scroll
CN109299373A (en) * 2018-10-20 2019-02-01 上海交通大学 Recommender system based on figure convolution technique
US20190362407A1 (en) * 2018-05-24 2019-11-28 International Business Machines Corporation Bundling items based on cost and value
CN110674407A (en) * 2019-09-30 2020-01-10 杭州电子科技大学 Hybrid recommendation method based on graph convolution neural network

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108713212A (en) * 2016-03-04 2018-10-26 范德曼公司 Method, system and medium for providing information based on grouping information
US20190362407A1 (en) * 2018-05-24 2019-11-28 International Business Machines Corporation Bundling items based on cost and value
CN108874914A (en) * 2018-05-29 2018-11-23 吉林大学 A kind of information recommendation method based on the long-pending and neural collaborative filtering of picture scroll
CN109299373A (en) * 2018-10-20 2019-02-01 上海交通大学 Recommender system based on figure convolution technique
CN110674407A (en) * 2019-09-30 2020-01-10 杭州电子科技大学 Hybrid recommendation method based on graph convolution neural network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
黄立威;江碧涛;吕守业;刘艳博;李德毅;: "基于深度学习的推荐系统研究综述", 计算机学报 *

Cited By (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112084407A (en) * 2020-09-08 2020-12-15 辽宁工程技术大学 Collaborative filtering recommendation method fusing graph neural network and attention mechanism
CN112084407B (en) * 2020-09-08 2024-03-12 辽宁工程技术大学 Collaborative filtering recommendation method integrating graph neural network and attention mechanism
CN112085171B (en) * 2020-09-15 2024-03-29 中山大学 Recommendation method based on clustered multi-entity graph neural network
CN112085171A (en) * 2020-09-15 2020-12-15 中山大学 Recommendation method based on clustering multi-entity graph neural network
CN112465593B (en) * 2020-11-27 2024-03-29 中国科学技术大学 Method for realizing fashion suit recommendation through graphic neural network
CN112465593A (en) * 2020-11-27 2021-03-09 中国科学技术大学 Method for realizing fashion suit recommendation through graph neural network
CN112651487A (en) * 2020-12-21 2021-04-13 广东交通职业技术学院 Data recommendation method, system and medium based on graph collapse convolution neural network
CN112651487B (en) * 2020-12-21 2021-07-27 广东交通职业技术学院 Data recommendation method, system and medium based on graph collapse convolution neural network
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
CN112650944A (en) * 2020-12-29 2021-04-13 哈尔滨理工大学 Collaborative filtering recommendation algorithm based on high-order connection graph
CN112686147B (en) * 2020-12-30 2024-05-28 北京纵目安驰智能科技有限公司 Vehicle and wheel subordinate relation prediction method, system, storage medium and terminal
CN112686147A (en) * 2020-12-30 2021-04-20 北京纵目安驰智能科技有限公司 Vehicle and wheel dependency prediction method and system, storage medium and terminal
CN113763014A (en) * 2021-01-05 2021-12-07 北京沃东天骏信息技术有限公司 Article co-occurrence relation determining method and device and judgment model obtaining method and device
CN112765373A (en) * 2021-01-29 2021-05-07 北京达佳互联信息技术有限公司 Resource recommendation method and device, electronic equipment and storage medium
CN112765373B (en) * 2021-01-29 2023-03-21 北京达佳互联信息技术有限公司 Resource recommendation method and device, electronic equipment and storage medium
CN112925977A (en) * 2021-02-26 2021-06-08 中国科学技术大学 Recommendation method based on self-supervision graph representation learning
CN113065045A (en) * 2021-04-20 2021-07-02 支付宝(杭州)信息技术有限公司 Method and device for carrying out crowd division and training multitask model on user
CN113177159A (en) * 2021-05-11 2021-07-27 清华大学 Binding recommendation method based on multichannel hypergraph neural network
CN113343121A (en) * 2021-06-02 2021-09-03 合肥工业大学 Lightweight graph convolution collaborative filtering recommendation method based on multi-granularity popularity characteristics
CN113343121B (en) * 2021-06-02 2022-08-09 合肥工业大学 Lightweight graph convolution collaborative filtering recommendation method based on multi-granularity popularity characteristics
CN113468227B (en) * 2021-06-25 2024-05-24 北京达佳互联信息技术有限公司 Information recommendation method, system, equipment and storage medium based on graph neural network
CN113468227A (en) * 2021-06-25 2021-10-01 北京达佳互联信息技术有限公司 Information recommendation method, system, device and storage medium based on graph neural network
CN114065048A (en) * 2021-11-25 2022-02-18 西安电子科技大学 Article recommendation method based on multi-different-pattern neural network
CN114065048B (en) * 2021-11-25 2024-02-09 西安电子科技大学 Article recommendation method based on multi-different composition graph neural network
CN114329200B (en) * 2021-12-27 2023-05-26 黑龙江大学 Personalized self-adaptive binding recommendation model and recommendation method
CN114329200A (en) * 2021-12-27 2022-04-12 黑龙江大学 Personalized self-adaptive binding recommendation model and recommendation method
CN114399028A (en) * 2022-01-14 2022-04-26 马上消费金融股份有限公司 Information processing method, graph convolution neural network training method and electronic equipment
CN116611894B (en) * 2023-06-02 2023-11-17 黑龙江大学 Cross-level graph comparison learning binding commodity recommendation method
CN116611894A (en) * 2023-06-02 2023-08-18 黑龙江大学 Cross-level graph comparison learning binding commodity recommendation method

Also Published As

Publication number Publication date
CN111611472B (en) 2024-02-27

Similar Documents

Publication Publication Date Title
CN111611472B (en) Binding recommendation method and system based on graph convolution neural network
US20200257976A1 (en) Algorithmic apparel recommendation
CN114036406A (en) Recommendation method and system based on graph contrast learning and social network enhancement
US20210342701A1 (en) Deep learning based visual compatibility prediction for bundle recommendations
CN113761388B (en) Recommendation method and device, electronic equipment and storage medium
CN116601626A (en) Personal knowledge graph construction method and device and related equipment
CN113761359B (en) Data packet recommendation method, device, electronic equipment and storage medium
CN113379494A (en) Commodity recommendation method and device based on heterogeneous social relationship and electronic equipment
CN112380453A (en) Article recommendation method and device, storage medium and equipment
CN113918834B (en) Graph convolution collaborative filtering recommendation method fusing social relations
CN110321473A (en) Diversity preference information method for pushing, system, medium and equipment based on multi-modal attention
CN112364242A (en) Graph convolution recommendation system for context-aware type
US20230267317A1 (en) Sign-aware recommendation apparatus and method using graph neural network
CN111966916A (en) Recommendation method and device, electronic equipment and computer readable storage medium
CN114579852A (en) Recommendation method based on graph convolution network and attention mechanism
CN112989182B (en) Information processing method, information processing device, information processing apparatus, and storage medium
WO2024061073A1 (en) Multimedia information generation method and apparatus, and computer-readable storage medium
US20240037133A1 (en) Method and apparatus for recommending cold start object, computer device, and storage medium
CN116662678A (en) Social interest embedding method adopting heterogeneous graph neural network
CN116541613A (en) Item recommendation method of double-message propagation mechanism based on scoring weighting
CN114936901A (en) Visual perception recommendation method and system based on cross-modal semantic reasoning and fusion
CN115221410A (en) Recommendation method and system based on desmooth graph convolution neural network
CN114611015A (en) Interactive information processing method and device and cloud server
CN114519600A (en) Graph neural network CTR estimation algorithm fusing adjacent node variances
CN113449193A (en) Information recommendation method and device based on multi-classification images

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