User attribute inference method and device based on hierarchical multi-channel hypergraph modeling

Abstract

The application relates to a user attribute inference method and device based on hierarchical multi-channel hypergraph modeling, wherein the method comprises the following steps: constructing hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups of users and items based on network data; constructing hypergraph pairs based on hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups fusion based on an attention mechanism; constructing a hierarchical multi-channel hypergraph convolution network according to the hypergraph pairs and corresponding hypergraph association matrices, and learning hypergraph embedding; and calculating user attributes according to overall representation and matrix decomposition, and optimizing a collaborative filtering task and a user attribute inference task. Thus, the problems in the related art that the collaborative filtering and the user attribute modeling inference task are not alternately optimized and mutually enhanced, the association in the user-item interaction network with attribute data is complex, it is difficult to accurately model high-order complex association therein, the attribute data and the interaction data are relatively sparse, and the user attribute inference effect is reduced are solved.

Description

Technical Field

[0001] This application relates to the fields of machine learning and user modeling technology, and in particular to a method and apparatus for inferring user attributes based on hierarchical multi-channel hypergraph modeling. Background Technology

[0002] With the rapid development of the information age, people are faced with all kinds of complex information every day, leading to the problem of information overload. It is becoming increasingly difficult for people to quickly and effectively obtain the information they need from the vast sea of ​​data. At the same time, different people often exhibit different preferences for different things, making it impossible to provide the same service to everyone. In this context, how to conduct personalized user modeling, obtain user attributes, and then provide personalized services such as personalized recommendations has become an urgent problem to be solved.

[0003] In related technologies, users' personalized preferences can be captured based on historical behavioral data between users and things, thereby enabling user modeling (i.e., collaborative filtering).

[0004] However, in related technologies, collaborative filtering and user attribute modeling and inference tasks are not optimized alternately and mutually reinforced. Furthermore, the relationships in user-item interaction networks with attribute data are complex, making it difficult to accurately model high-order complex relationships. Both attribute data and interaction data are relatively sparse, which reduces the effectiveness of user attribute inference and urgently needs improvement. Summary of the Invention

[0005] This application provides a user attribute inference method and apparatus based on hierarchical multi-channel hypergraph modeling to solve the problems in related technologies, such as the failure to alternately optimize and mutually enhance collaborative filtering and user attribute modeling and inference tasks, the complexity of associations in user-item interaction networks with attribute data, the difficulty in accurately modeling high-order complex associations, and the sparseness of attribute data and interaction data, which reduces the effectiveness of user attribute inference.

[0006] The first aspect of this application provides a user attribute inference method based on hierarchical multi-channel hypergraph modeling, comprising the following steps: processing user-item interaction network and attribute data into network data; constructing hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups for users and items based on the network data; fusing the hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups based on an attention mechanism to construct hypergraph pairs, and obtaining the hypergraph association matrix corresponding to the hypergraph pairs; constructing a hierarchical multi-channel hypergraph convolutional network according to the hypergraph pairs and the corresponding hypergraph association matrix, and learning hypergraph embeddings to learn user and item attribute space representations and interaction space representations, and learning the semantic space representation of each attribute based on fully connected layers; fusing the learned user, item attribute space representations and interaction space representations into an overall representation, and predicting user-item interaction scores based on inner products; inferring user attributes based on the overall representation and matrix factorization, and optimizing collaborative filtering tasks and user attribute inference tasks.

[0007] Optionally, in one embodiment of this application, it further includes: unsupervisedly narrowing the distance between user and item attribute representations and attribute space semantic representations based on contrastive learning in self-supervised learning.

[0008] Optionally, in one embodiment of this application, the construction of hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups based on the network data includes: for a target attribute, constructing a hyperedge connecting all nodes having the target attribute, and for each sub-attribute of the target attribute, constructing a hyperedge connecting all nodes having the sub-attribute, to construct a hierarchical hyperedge group; combining attributes, wherein a hyperedge is constructed connecting all nodes simultaneously having the attribute combination; for the interaction hyperedge group, it is constructed based on the first-order nearest neighbors of users and items, wherein when the user is a hyperedge, it connects all items that the user has interacted with, and when the item is a hyperedge, it connects all users that have interacted with the item.

[0009] Optionally, in one embodiment of this application, the hierarchical multi-channel hypergraph convolutional network has 2 layers and the hidden layer feature dimension is 32.

[0010] Optionally, in one embodiment of this application, the optimization of the collaborative filtering task and the user attribute inference task includes: jointly optimizing the collaborative filtering task and the user attribute inference task based on Bayesian personalized pairwise loss and multi-channel cross-entropy loss.

[0011] A second aspect of this application provides a user attribute inference apparatus based on hierarchical multi-channel hypergraph modeling, comprising: a processing module for processing user-item interaction network and attribute data into network data; a construction module for constructing hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups for users and items based on the network data; an acquisition module for fusing the hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups based on an attention mechanism to construct hypergraph pairs and acquire the hypergraph association matrices corresponding to the hypergraph pairs; a learning module for constructing a hierarchical multi-channel hypergraph convolutional network based on the hypergraph pairs and the corresponding hypergraph association matrices, and learning hypergraph embeddings to learn user and item attribute space representations and interaction space representations, and learning the semantic space representation of each attribute based on fully connected layers; a prediction module for fusing the learned user, item attribute space representations and interaction space representations into an overall representation, and predicting user-item interaction scores based on inner products; and an inference module for inferring user attributes based on the overall representation and matrix factorization, and optimizing collaborative filtering tasks and user attribute inference tasks.

[0012] Optionally, in one embodiment of this application, it further includes: a control module, used to unsupervisedly narrow the distance between the user and item attribute representations and the attribute space semantic representations based on the contrastive learning method in self-supervised learning.

[0013] Optionally, in one embodiment of this application, the construction module includes: a first construction unit, configured to construct a hyperedge connecting all nodes having the target attribute for a target attribute, and to construct hyperedges connecting all nodes having the sub-attributes of the target attribute, respectively, to construct a hierarchical hyperedge group; a second construction unit, configured to combine attributes, wherein a hyperedge is constructed connecting all nodes simultaneously having the attribute combination; and a connection unit, configured to construct the interaction hyperedge group based on the first-order nearest neighbors of users and items, wherein when the user is a hyperedge, all items interacted by the user are connected, and when the item is a hyperedge, all users interacted with the item are connected.

[0014] Optionally, in one embodiment of this application, the hierarchical multi-channel hypergraph convolutional network has 2 layers and the hidden layer feature dimension is 32.

[0015] Optionally, in one embodiment of this application, the inference module includes: an optimization unit, used to jointly optimize the collaborative filtering task and the user attribute inference task based on Bayesian personalized pairwise loss and multi-channel cross-entropy loss.

[0016] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the user attribute inference method based on hierarchical multi-channel hypergraph modeling as described in the above embodiments.

[0017] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described user attribute inference method based on hierarchical multichannel hypergraph modeling.

[0018] This application's embodiments can model high-order complex relationships in user-item networks with attribute data based on hierarchical multi-channel hypergraphs, learn accurate node embeddings, and perform user attribute inference, thereby improving the performance of user attribute inference methods and accurately improving the modeling of high-order relationships such as attributes. This solves the problems in related technologies, such as the failure to alternately optimize and mutually reinforce collaborative filtering and user attribute modeling and inference tasks, the complexity of relationships in user-item interaction networks with attribute data making it difficult to accurately model high-order complex relationships, and the sparseness of both attribute and interaction data reducing the effectiveness of user attribute inference.

[0019] Additional aspects and advantages of this 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 this application. Attached Figure Description

[0020] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0021] Figure 1 This is a flowchart illustrating a user attribute inference method based on hierarchical multichannel hypergraph modeling according to an embodiment of this application;

[0022] Figure 2 This is a flowchart of a user attribute inference method based on hierarchical multichannel hypergraph modeling according to an embodiment of this application;

[0023] Figure 3 This is a schematic diagram of a user attribute inference device based on hierarchical multichannel hypergraph modeling provided in an embodiment of this application;

[0024] Figure 4 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0025] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0026] The following description, with reference to the accompanying drawings, outlines a user attribute inference method and apparatus based on hierarchical multi-channel hypergraph modeling, according to embodiments of this application. Addressing the issues raised in the background section regarding the related technologies where collaborative filtering and user attribute modeling inference tasks are not alternately optimized and mutually reinforced, and where complex associations in user-item interaction networks with attribute data make it difficult to accurately model high-order complex associations, and where both attribute and interaction data are sparse, thus reducing the effectiveness of user attribute inference, this application provides a user attribute inference method based on hierarchical multi-channel hypergraph modeling. This method can model high-order complex associations in user-item networks with attribute data using hierarchical multi-channel hypergraph modeling, learn accurate node embeddings, and perform user attribute inference, thereby improving the performance of the user attribute inference method and accurately improving the modeling of high-order associations such as attributes. This solves the problems in the related technologies where collaborative filtering and user attribute modeling inference tasks are not alternately optimized and mutually reinforced, and where complex associations in user-item interaction networks with attribute data make it difficult to accurately model high-order complex associations, and where both attribute and interaction data are sparse, thus reducing the effectiveness of user attribute inference.

[0027] Specifically, Figure 1 This is a flowchart illustrating a user attribute inference method based on hierarchical multi-channel hypergraph modeling provided in an embodiment of this application.

[0028] like Figure 1 As shown, the user attribute inference method based on hierarchical multi-channel hypergraph modeling includes the following steps:

[0029] In step S101, the user-item interaction network and attribute data are processed into network data.

[0030] It is understandable that in this embodiment of the application, the accuracy of user modeling is greatly affected when historical user behavior is lacking. User attribute information, such as gender, age, and occupation, can assist user modeling to a certain extent. The user attribute information in this embodiment of the application is very important for personalized user modeling because users with similar characteristics often have similar preferences. Similarly, things with the same characteristics will attract similar users. Based on attribute information, the similarity between users or things can be modeled more accurately and effectively.

[0031] In actual implementation, this application embodiment performs user attribute inference on user-item interaction data with attribute data. First, data preprocessing can be performed, converting the acquired user-item interaction network and attribute data into network data format, such as... This provides a basis for the subsequent construction of hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups, quickly establishes the relationship between users and things, ensures accurate node embedding and attribute representation for complex user-item interaction networks and attribute data, and improves the effect of user attribute inference.

[0032] In step S102, hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups for users and items are constructed based on network data.

[0033] In actual implementation, the embodiments of this application can construct hyperedge groups, and build hierarchical multi-channel attribute hyperedge groups for users and items based on data. and interactive hyper-edge groups Where n1 and n2 are the number of hierarchical multi-channel attribute hyperedge groups for users and items, respectively.

[0034] The embodiments of this application can construct hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups for users and items based on network data, thereby achieving accurate modeling of complex user-item interaction relationships with attribute data through hierarchical multi-channel hypergraph modeling.

[0035] Optionally, in one embodiment of this application, constructing hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups for users and items based on network data includes: for a target attribute, constructing a hyperedge connecting all nodes with the target attribute, and for sub-attributes of the target attribute, constructing hyperedges connecting all nodes with the sub-attributes, to construct a hierarchical hyperedge group; combining attributes, wherein a hyperedge is constructed connecting all nodes that simultaneously have attribute combinations; for interaction hyperedge groups, constructing them based on the first-order nearest neighbors of users and items, wherein when a user is a hyperedge, it connects all items that the user has interacted with, and when an item is a hyperedge, it connects all users that have interacted with the item.

[0036] In some embodiments, a hierarchical multi-channel hyperedge group can first be constructed. For example, for the target attribute a, a hyperedge can be constructed to connect all nodes with the target attribute. Based on this, for the sub-attributes a1, a2, etc. of the target attribute a, hyperedges can be constructed to connect all nodes with the sub-attributes, thereby constructing a hierarchical hyperedge group. In addition, the embodiments of this application can combine attributes. For example, for attributes a and attribute b, a hyperedge can be constructed to connect all nodes that simultaneously have the combination of attributes a and attribute b.

[0037] Furthermore, in this embodiment of the application, the interaction hyperedge group can be constructed based on the first-order nearest neighbors of users and items. When a user is a hyperedge, it connects all items that the user has interacted with, and when an item is a hyperedge, it connects all users that have interacted with the item. This enables accurate modeling of complex user-item interaction relationships with attribute data.

[0038] Optionally, in one embodiment of this application, the hierarchical multi-channel hypergraph convolutional network has 2 layers and the hidden layer feature dimension is 32.

[0039] In some cases, embodiments of this application can construct a hierarchical multi-channel hypergraph convolutional network suitable for user-item interaction data with attributes. The network has two layers, with 32-dimensional hidden layer features, and learns hypergraph embeddings. and This solves the problem of accurately modeling high-order relationships such as attributes.

[0040] In step S103, the hierarchical multi-channel attribute hyperedge group and interaction hyperedge group based on the attention mechanism are fused to construct a hypergraph pair, and the hypergraph association matrix corresponding to the hypergraph pair is obtained.

[0041] In actual implementation, embodiments of this application can construct hypergraphs, and construct hypergraph pairs by fusing attribute hyperedge groups and interaction hyperedge groups based on attention mechanisms. and and Constructed in a similar manner, embodiments of this application can obtain the hypergraph association matrix corresponding to hypergraph pairs. and Where Atten(·) is the attention function, It is represented as a learnable weight matrix for the user in the interaction space.

[0042] The embodiments of this application can construct hypergraph pairs by fusing hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups based on attention mechanisms, and obtain the hypergraph association matrix corresponding to the hypergraph pairs, thereby achieving accurate modeling of complex interactive associations.

[0043] In step S104, a hierarchical multi-channel hypergraph convolutional network is constructed based on the hypergraph pairs and the corresponding hypergraph association matrices, and hypergraph embeddings are learned to learn the user, item attribute space representation and interaction space representation, and the semantic space representation of each attribute is learned based on the fully connected layer.

[0044] In some cases, embodiments of this application can construct a hierarchical multi-channel hypergraph convolutional network based on hypergraph pairs and their corresponding hypergraph association matrices to learn hypergraph embeddings. and Constructing a hierarchical multichannel hypergraph convolutional network suitable for user-item interaction data with attributes:

[0045]

[0046]

[0047] in, and Let U and V be the node features in the attribute spaces of the node sets U and V in the l-th layer, respectively. and Let θ represent the node features in the interaction space of node sets U and V in the l-th layer, respectively, and θ be a learnable parameter.

[0048] After performing L-layer hierarchical multi-channel hypergraph convolution, the embodiments of this application can obtain node features. and

[0049] Furthermore, embodiments of this application can learn user and item attribute space representations and interaction space representations, such as performing attribute representation learning and initializing attribute representations a1, a2, ..., a n+m It learns, and learns the semantic space representation of each attribute based on the fully connected layer.

[0050] The embodiments of this application can construct a hierarchical multi-channel hypergraph convolutional network and learn hypergraph embeddings to learn user and item attribute space representations and interaction space representations. In this way, the hierarchical multi-channel hypergraph convolutional network can learn the semantic representation of attributes as well as the attribute and interaction space representations of users and items, making full use of the supplementary information of attributes and improving the effect of user attribute inference.

[0051] In step S105, the learned user and item attribute space representations and interaction space representations are fused into an overall representation, and the user-item interaction score is predicted based on the inner product.

[0052] In practical implementation, the embodiments of this application can perform collaborative filtering by constructing a hierarchical multi-channel hypergraph convolutional network and learning hypergraph embeddings, thereby fusing the learned user and item attribute space representations and interaction space representations into an overall representation. and This application embodiment can predict the user-item interaction score R through the inner product: R = X U (X V ) T This further improves the effectiveness of user attribute inference.

[0053] In step S106, user attributes are inferred based on the overall representation and matrix factorization, and the collaborative filtering task and user attribute inference task are optimized.

[0054] As one possible implementation, embodiments of this application can perform user attribute inference based on the overall representation and matrix decomposition: A = X U P T B = X V K T To infer user attributes The embodiments of this application can optimize collaborative filtering tasks and user attribute inference tasks, thereby improving the effect of user attribute inference.

[0055] Optionally, in one embodiment of this application, optimizing the collaborative filtering task and the user attribute inference task includes: jointly optimizing the collaborative filtering task and the user attribute inference task based on Bayesian personalized pairwise loss and multi-channel cross-entropy loss.

[0056] In actual implementation, the embodiments of this application can perform multi-task alternation optimization, using Bayesian Personalized Ranking (BPR):

[0057]

[0058] Multi-channel CrossEntropy (MCE):

[0059]

[0060] The embodiments of this application can jointly optimize collaborative filtering and user attribute inference tasks using Bayesian Personalized Ranking (BPR) and Multi-channel CrossEntropy (MCE) loss, thereby further improving the performance of user attribute inference methods and increasing the accuracy of modeled attributes.

[0061] Optionally, in one embodiment of this application, it further includes: unsupervisedly narrowing the distance between user and item attribute representations and attribute space semantic representations based on contrastive learning in self-supervised learning.

[0062] In other embodiments, self-supervised enhanced attribute learning inference can be performed. Based on the contrastive learning paradigm in self-supervised learning, the distance between user and item attribute representations and the semantic representations in the attribute space is unsupervisedly narrowed, enhancing representational capabilities. Contrastive learning in the attribute space optimizes a custom loss function.

[0063] express.

[0064] The embodiments of this application can unsupervisedly narrow the distance between user and item attribute representations and attribute space semantic representations based on the contrastive learning method in self-supervised learning, thereby achieving accurate modeling of complex user-item interaction relationships with attribute data.

[0065] Specifically, in combination Figure 2 As shown, a specific embodiment is used to elaborate in detail on the working principle of the user attribute inference method based on hierarchical multi-channel hypergraph modeling in this application.

[0066] like Figure 2 As shown, embodiments of this application may include the following steps:

[0067] Step S201: Data preprocessing. This embodiment of the application can perform data preprocessing, converting the acquired user-item interaction network and attribute data into network data.

[0068] Step S202: Construct hyperedge groups. Embodiments of this application can construct hyperedge groups, including hierarchical multi-channel attribute hyperedge groups and interactive hyperedge groups, based on the data.

[0069] Step S203: Construct a hypergraph. This embodiment of the application can construct a hypergraph by fusing attribute hyperedge groups and interaction hyperedge groups based on the attention mechanism.

[0070] Step S204: Construct a hierarchical multi-channel hypergraph convolutional network. This embodiment of the application can construct a hierarchical multi-channel hypergraph convolutional network to learn hypergraph embeddings.

[0071] Step S205: Attribute Representation Learning. This embodiment of the application can learn attribute representations by initializing and learning the semantic space representation of each attribute based on the fully connected layer.

[0072] Step S206: Collaborative Filtering. Embodiments of this application can use collaborative filtering to predict user-item interactions based on the inner product of user and item representations.

[0073] Step S207: User attribute inference. This embodiment of the application can perform user attribute inference, inferring user attributes based on attribute representation and matrix decomposition.

[0074] Step S208: Multi-task alternating optimization. This embodiment of the application can perform multi-task alternating optimization, jointly optimizing collaborative filtering tasks and user attribute inference tasks.

[0075] Step S209: Self-supervised enhanced attribute learning inference. This embodiment of the application can perform self-supervised enhanced attribute learning inference, further enhancing the attribute learning inference based on the contrastive learning paradigm in self-supervised learning.

[0076] The user attribute inference method based on hierarchical multi-channel hypergraph modeling proposed in this application can model high-order complex relationships in user-item networks with attribute data using hierarchical multi-channel hypergraph modeling. It learns accurate node embeddings and performs user attribute inference, thereby improving the performance of the user attribute inference method and accurately improving the modeling of high-order relationships such as attributes. This solves the problems in related technologies where collaborative filtering and user attribute modeling and inference tasks are not alternately optimized and mutually reinforced, and where the relationships in user-item interaction networks with attribute data are complex, making it difficult to accurately model high-order complex relationships. Furthermore, the sparse attribute and interaction data reduce the effectiveness of user attribute inference.

[0077] Next, with reference to the accompanying drawings, a user attribute inference apparatus based on hierarchical multi-channel hypergraph modeling proposed in this application is described.

[0078] Figure 3 This is a schematic diagram of the user attribute inference device based on hierarchical multi-channel hypergraph modeling according to an embodiment of this application.

[0079] like Figure 3 As shown, the user attribute inference device 10 based on hierarchical multi-channel hypergraph modeling includes: a processing module 100, a construction module 200, an acquisition module 300, a learning module 400, a prediction module 500, and an inference module 600.

[0080] Specifically, the processing module 100 is used to process the user-item interaction network and attribute data into network data.

[0081] Module 200 is used to construct hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups for users and items based on network data.

[0082] The module 300 is used to construct hypergraph pairs by fusing hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups based on the attention mechanism, and to obtain the hypergraph association matrix corresponding to the hypergraph pairs.

[0083] The learning module 400 is used to construct a hierarchical multi-channel hypergraph convolutional network based on hypergraph pairs and their corresponding hypergraph association matrices, and to learn hypergraph embeddings to learn user and item attribute space representations and interaction space representations, and to learn the semantic space representation of each attribute based on fully connected layers.

[0084] The prediction module 500 is used to merge the learned user and item attribute space representations and interaction space representations into an overall representation, and predict user-item interaction scores based on the inner product.

[0085] The inference module 600 is used to infer user attributes based on the overall representation and matrix factorization, and to optimize collaborative filtering and user attribute inference tasks.

[0086] Optionally, in one embodiment of this application, the user attribute inference device 10 based on hierarchical multi-channel hypergraph modeling further includes a control module.

[0087] Among them, the control module is used to unsupervisedly narrow the distance between the user and item attribute representations and the attribute space semantic representations based on the contrastive learning method in self-supervised learning.

[0088] Optionally, in one embodiment of this application, the building module 200 includes: a first building unit, a second building unit, and a connection unit.

[0089] The first construction unit is used to construct a hyperedge connecting all nodes with the target attribute for the target attribute, and to construct hyperedges connecting all nodes with the sub-attributes of the target attribute for the sub-attributes, so as to construct a hierarchical hyperedge group.

[0090] The second construction unit is used to combine attributes, wherein a hyperedge is constructed to connect all nodes that simultaneously have attribute combinations.

[0091] A connection unit is used to construct a group of interactive hyperedges based on the first-order nearest neighbors of users and items. When a user is a hyperedge, it connects all items that the user has interacted with, and when an item is a hyperedge, it connects all users that have interacted with the item.

[0092] Optionally, in one embodiment of this application, the hierarchical multi-channel hypergraph convolutional network has 2 layers and the hidden layer feature dimension is 32.

[0093] Optionally, in one embodiment of this application, the inference module 600 includes an optimization unit.

[0094] The optimization unit is used to jointly optimize collaborative filtering and user attribute inference tasks based on Bayesian personalized pairwise loss and multi-channel cross-entropy loss.

[0095] It should be noted that the explanation of the aforementioned user attribute inference method embodiment for hierarchical multichannel hypergraph modeling also applies to the user attribute inference device for hierarchical multichannel hypergraph modeling in this embodiment, and will not be repeated here.

[0096] The user attribute inference device based on hierarchical multi-channel hypergraph modeling proposed in this application can model high-order complex relationships in user-item networks with attribute data based on hierarchical multi-channel hypergraph modeling, learn accurate node embeddings, and perform user attribute inference, thereby improving the performance of user attribute inference methods and accurately improving the modeling of high-order relationships such as attributes. This solves the problems in related technologies where collaborative filtering and user attribute modeling and inference tasks are not alternately optimized and mutually reinforced, and where the relationships in user-item interaction networks with attribute data are complex, making it difficult to accurately model high-order complex relationships, and where both attribute data and interaction data are relatively sparse, thus reducing the effectiveness of user attribute inference.

[0097] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0098] The memory 401, the processor 402, and the computer program stored on the memory 401 and capable of running on the processor 402.

[0099] When the processor 402 executes the program, it implements the user attribute inference method based on hierarchical multi-channel hypergraph modeling provided in the above embodiments.

[0100] Furthermore, electronic devices also include:

[0101] Communication interface 403 is used for communication between memory 401 and processor 402.

[0102] The memory 401 is used to store computer programs that can run on the processor 402.

[0103] The memory 401 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0104] If the memory 401, processor 402, and communication interface 403 are implemented independently, then the communication interface 403, memory 401, and processor 402 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be divided into address buses, data buses, control buses, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0105] Optionally, in a specific implementation, if the memory 401, processor 402, and communication interface 403 are integrated on a single chip, then the memory 401, processor 402, and communication interface 403 can communicate with each other through an internal interface.

[0106] Processor 402 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0107] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described user attribute inference method based on hierarchical multichannel hypergraph modeling.

[0108] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0109] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0110] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0111] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0112] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0113] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0114] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0115] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A user attribute inference method based on hierarchical multi-channel hypergraph modeling, characterized in that, Includes the following steps: Process user-item interaction network and attribute data into network data; Based on the network data, hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups for users and items are constructed; The hierarchical multi-channel attribute hyperedge group and the interaction hyperedge group based on the attention mechanism are fused to construct a hypergraph pair, and the hypergraph association matrix corresponding to the hypergraph pair is obtained. A hierarchical multi-channel hypergraph convolutional network is constructed based on the hypergraph pairs and the corresponding hypergraph association matrices, and hypergraph embeddings are learned to learn user and item attribute space representations and interaction space representations. Furthermore, the semantic space representation of each attribute is learned based on fully connected layers. The learned user, item attribute space representation, and interaction space representation are merged into an overall representation, and the user-item interaction score is predicted based on the inner product. User attributes are inferred based on the overall representation and matrix factorization, and the collaborative filtering task and user attribute inference task are optimized.

2. The method according to claim 1, characterized in that, Also includes: The contrastive learning approach in self-supervised learning unsupervisedly bridges the gap between user and item attribute representations and attribute space semantic representations.

3. The method according to claim 1, characterized in that, The construction of hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups for users and items based on the network data includes: For the target attribute, construct a hyperedge connecting all nodes with the target attribute, and for the sub-attributes of the target attribute, construct hyperedges connecting all nodes with the sub-attributes respectively, to construct a hierarchical hyperedge group. The attributes are combined, wherein a hyperedge is constructed to connect all nodes that simultaneously possess the attribute combination; For interactive hyperedge groups, a first-order nearest neighbor structure based on users and items is constructed, wherein when a user is a hyperedge, it connects all items that the user has interacted with, and when an item is a hyperedge, it connects all users that have interacted with the item.

4. The method according to claim 1, characterized in that, The hierarchical multi-channel hypergraph convolutional network has 2 layers and the hidden layer features have a dimension of 32.

5. The method according to claim 1, characterized in that, The optimized collaborative filtering task and user attribute inference task include: The collaborative filtering task and the user attribute inference task are jointly optimized based on Bayesian personalized pairwise loss and multi-channel cross-entropy loss.

6. A user attribute inference device based on hierarchical multi-channel hypergraph modeling, characterized in that, include: The processing module is used to process user-item interaction network and attribute data into network data; The construction module is used to construct hierarchical multi-channel attribute hyperedge groups and interaction hyperedge groups for users and items based on the network data; The acquisition module is used to construct a hypergraph pair by fusing the hierarchical multi-channel attribute hyperedge group and the interaction hyperedge group based on the attention mechanism, and to acquire the hypergraph association matrix corresponding to the hypergraph pair; The learning module is used to construct a hierarchical multi-channel hypergraph convolutional network based on the hypergraph pairs and the corresponding hypergraph association matrices, and to learn hypergraph embeddings to learn user and item attribute space representations and interaction space representations, and to learn the semantic space representation of each attribute based on fully connected layers; The prediction module is used to fuse the learned user, the item attribute space representation, and the interaction space representation into an overall representation, and predict the user-item interaction score based on the inner product; The inference module is used to infer user attributes based on the overall representation and matrix decomposition, and to optimize the collaborative filtering task and the user attribute inference task.

7. The apparatus according to claim 6, characterized in that, Also includes: The control module is used to unsupervisedly narrow the gap between user and item attribute representations and attribute space semantic representations based on contrastive learning in self-supervised learning.

8. The apparatus according to claim 6, characterized in that, The building module includes: The first construction unit is used to construct a hyperedge connecting all nodes with the target attribute for the target attribute, and to construct hyperedges connecting all nodes with the sub-attributes of the target attribute for the sub-attributes, so as to construct a hierarchical hyperedge group. The second construction unit is used to combine attributes, wherein a hyperedge is constructed to connect all nodes that simultaneously have the attribute combination; A connection unit is used to construct, for example, an interactive hyperedge group based on the first-order nearest neighbors of users and items, wherein when the user is a hyperedge, it connects all items that the user has interacted with, and when an item is a hyperedge, it connects all users that have interacted with the item.

9. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, the processor executing the program to implement the user attribute inference method based on hierarchical multichannel hypergraph modeling as described in any one of claims 1-5.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the user attribute inference method based on hierarchical multichannel hypergraph modeling as described in any one of claims 1-5.

Images