Cross-domain sequence recommendation method and device based on hypergraph convolutional neural network

The local and global hypergraphs are constructed through hypergraph convolutional neural network, and the collaborative information of the timing subsequence of user interaction projects is extracted, which solves the problem of inaccurate project recommendation caused by short user behavior sequences or sparse data in the prior art, and improves the recommended Accuracy and user experience.

CN118939871AActive Publication Date: 2024-11-12ZHEJIANG UNIV

Patent Information

Application Number
CN202410811559.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-21
Publication Date
2024-11-12
Estimated Expiration
2044-06-21

AI Technical Summary

Technical Problem

When existing sequence recommendation models face short user behavior sequences or sparse data, it is difficult to accurately describe user personalized preferences, resulting in inaccurate project recommendations.

Method used

Local and global hypergraphs are constructed using hypergraph convolutional neural networks, and the collaboration information between and internal subsequences of user interaction projects is extracted through the cross-domain sequence recommendation model, and user project interaction characteristics are generated to predict the items to be interacted with.

Benefits of technology

It improves the accuracy of project recommendations and improves user experience. By considering the influence relationship between the same user and different users in a certain period of interaction, more rich information can be obtained.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118939871A_ABST
    Figure CN118939871A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides a cross-domain sequence recommendation method and device based on a hypergraph convolutional neural network. The cross-domain sequence recommendation method based on the hypergraph convolutional neural network comprises the steps that historical user interaction item sequential sequences of multiple domains are acquired; local hypergraphs and global hypergraphs corresponding to different time periods are constructed, the local hypergraphs are hypergraphs constructed based on interaction items in historical user interaction item sequential sequences of target domains in the corresponding time periods, and the global hypergraphs are hypergraphs constructed based on interaction items in historical user interaction item sequential sequences of multiple domains in the corresponding time periods; the target domain is any one of a plurality of domains, and any one hyperedge in the local hypergraph and the global hypergraph represents an interaction item time sequence sub-sequence formed by interaction items of any user in a corresponding time period; and inputting the local hypergraph and the global hypergraph into a cross-domain sequence recommendation model, and predicting a to-be-interacted item of the user. The item recommendation accuracy can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of this specification relate to the field of recommendation technology, and in particular to a cross-domain sequence recommendation method and device based on a hypergraph convolutional neural network. Background Art

[0002] Sequential recommendation is one of the core tasks in the field of recommendation systems. The core goal of sequential recommendation is to capture the interests that evolve dynamically over time by deeply analyzing the user's historical behavior sequence, and then provide users with accurate next item recommendations. When faced with situations where the user behavior sequence is short or the data is sparse, most existing sequential recommendation models still face considerable challenges in accurately depicting the user's personalized preferences.

[0003] In the existing sequence recommendation schemes, there is still the problem of inaccurate item recommendations. Therefore, how to improve the accuracy of item recommendations based on user historical behavior sequences is a problem that needs to be solved at present. Summary of the invention

[0004] In view of this, the embodiments of this specification provide a cross-domain sequence recommendation method of a hypergraph convolutional neural network. One or more embodiments of this specification also relate to a cross-domain sequence recommendation device based on a hypergraph convolutional neural network, a recommendation system, a computing device, a computer-readable storage medium, and a computer program product to solve the technical defects existing in the prior art.

[0005] According to a first aspect of an embodiment of this specification, a cross-domain sequence recommendation method based on a hypergraph convolutional neural network is provided, comprising:

[0006] Acquire a time series sequence of historical user interaction items in multiple domains, wherein the time series sequence of historical user interaction items includes interaction items corresponding to multiple users; construct a local hypergraph and a global hypergraph corresponding to different time periods, wherein the local hypergraph is a hypergraph constructed based on the interaction items in the time series sequence of historical user interaction items in the target domain within the corresponding time period, and the global hypergraph is a hypergraph constructed based on the interaction items in the time series sequence of historical user interaction items in the multiple domains within the corresponding time period, the target domain is any domain among the multiple domains, and any hyperedge in the local hypergraph and the global hypergraph represents an interaction item constituted by the interaction items of any user in the corresponding time period. temporal subsequence; inputting the local hypergraph and the global hypergraph into a cross-domain sequence recommendation model to predict the user's items to be interacted; wherein the cross-domain sequence recommendation model is used to extract the collaboration information between the interactive item temporal subsequences of different users and within the interactive item temporal subsequences based on the local hypergraph and the global hypergraph through a hypergraph convolutional network, obtain the item features of the interactive items containing the collaboration information, determine the user-item interaction features of any user based on the item features corresponding to the interactive items in the interactive item temporal subsequence of multiple time periods, and predict the items to be interacted based on the user-item interaction features.

[0007] According to a second aspect of an embodiment of this specification, a cross-domain sequence recommendation model training method is provided, including:

[0008] Obtain user interaction project time series samples of multiple domains and label information corresponding to the user interaction project time series samples from a training sample set, wherein the user interaction project time series samples include interaction projects corresponding to multiple users; construct local hypergraphs and global hypergraphs corresponding to different time periods, wherein the local hypergraph is a hypergraph constructed based on the interaction projects in the user interaction project time series samples of the target domain within the corresponding time period, and the global hypergraph is a hypergraph constructed based on the interaction projects in the user interaction project time series samples of the multiple domains within the corresponding time period, the target domain is any domain among the multiple domains, and any hyperedge in the local hypergraph and the global hypergraph represents an interaction project time series subsequence composed of interaction projects of any user within the corresponding time period; input the local hypergraph and the global hypergraph into a preset model to predict the user's items to be interacted, wherein the preset model is used to predict the user's items to be interacted based on the local hypergraph through a hypergraph convolutional network. Hypergraph and global hypergraph, extract the collaborative information between the interactive project time series subsequences of different users and within the interactive project time series subsequences, obtain the project features of the interactive projects containing the collaborative information, determine the user-project interaction features of any user based on the project features corresponding to the interactive projects in the interactive project time series subsequences of multiple time periods, and predict the items to be interacted based on the user-project interaction features; calculate the training loss using the items to be interacted and the label information; update the parameters of the preset model based on the training loss to obtain the preset model after the parameters are updated, if the training end condition is not met, continue to execute the step of obtaining the user interaction project time series sequence samples of multiple domains and the label information corresponding to the user interaction project time series sequence samples from the training sample set, when the training end condition is met, determine the current preset model after the parameters are updated as the cross-domain sequence recommendation model.

[0009] According to a third aspect of the embodiments of this specification, a cross-domain sequence recommendation device based on a hypergraph convolutional neural network is provided, comprising:

[0010] A sequence acquisition module, used to acquire a time sequence sequence of historical user interaction items of multiple domains, wherein the time sequence sequence of historical user interaction items includes interaction items corresponding to multiple users;

[0011] A hypergraph construction module, used to construct local hypergraphs and global hypergraphs corresponding to different time periods, wherein the local hypergraph is a hypergraph constructed based on the interaction items in the historical user interaction item time sequence of the target domain in the corresponding time period, and the global hypergraph is a hypergraph constructed based on the interaction items in the historical user interaction item time sequence of the multiple domains in the corresponding time period, the target domain is any domain among the multiple domains, and any hyperedge in the local hypergraph and the global hypergraph represents an interaction item time sequence subsequence composed of the interaction items of any user in the corresponding time period;

[0012] An item prediction module, used to input the local hypergraph and the global hypergraph into a cross-domain sequence recommendation model to predict items to be interacted with by the user;

[0013] Among them, the cross-domain sequence recommendation model is used to extract the collaborative information between the interactive project time subsequences of different users and within the interactive project time subsequences based on the local hypergraph and the global hypergraph through a hypergraph convolutional network, obtain the project features of the interactive projects containing the collaborative information, determine the user-project interaction features of any user based on the project features corresponding to the interactive projects in the interactive project time subsequences of multiple time periods, and predict the project to be interacted based on the user-project interaction features.

[0014] According to a fourth aspect of the embodiments of this specification, a recommendation system is provided, including a server and a client device, wherein:

[0015] The server is used to: obtain a time sequence of historical user interaction items in multiple domains, wherein the time sequence of historical user interaction items includes interaction items corresponding to multiple users; construct a local hypergraph and a global hypergraph corresponding to different time periods, wherein the local hypergraph is a hypergraph constructed based on interaction items in the time sequence of historical user interaction items in a target domain within a corresponding time period, and the global hypergraph is a hypergraph constructed based on interaction items in the time sequence of historical user interaction items in the multiple domains within a corresponding time period, wherein the target domain is any domain among the multiple domains, and any hyperedge in the local hypergraph and the global hypergraph represents an interaction item time sequence subsequence constituted by interaction items of any user within the corresponding time period; and The local hypergraph and the global hypergraph are input into a cross-domain sequence recommendation model to predict the user's to-be-interacted items; the recommendation information corresponding to the to-be-interacted items is sent to the client device corresponding to the user, and the cross-domain sequence recommendation model is used to extract the collaboration information between the interactive item time subsequences of different users and within the interactive item time subsequences based on the local hypergraph and the global hypergraph through a hypergraph convolutional network, obtain the item features of the interactive items containing the collaboration information, determine the user-item interaction features of the user based on the item features corresponding to the interactive items in the interactive item time subsequence of any user in multiple time periods, and predict the to-be-interacted items based on the user-item interaction features;

[0016] The client device is used to receive the recommendation information and display the items to be interacted with through a user interaction interface.

[0017] According to a fifth aspect of an embodiment of this specification, a computing device is provided, including:

[0018] Memory and processor;

[0019] The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer program / instructions are executed by the processor, the steps of the cross-domain sequence recommendation method based on the hypergraph convolutional neural network and / or the cross-domain sequence recommendation model training method are implemented.

[0020] According to the sixth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned cross-domain sequence recommendation method based on a hypergraph convolutional neural network, and / or the above-mentioned cross-domain sequence recommendation model training method.

[0021] According to the seventh aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned cross-domain sequence recommendation method based on a hypergraph convolutional neural network, and / or the above-mentioned cross-domain sequence recommendation model training method.

[0022] An embodiment of the present specification implements an information extraction method, including: obtaining a time series sequence of historical user interaction items in multiple domains, wherein the time series sequence of historical user interaction items includes interaction items corresponding to multiple users; constructing a local hypergraph and a global hypergraph corresponding to different time periods, wherein the local hypergraph is a hypergraph constructed based on the interaction items in the time series sequence of historical user interaction items in a target domain within the corresponding time period, and the global hypergraph is a hypergraph constructed based on the interaction items in the time series sequence of historical user interaction items in the multiple domains within the corresponding time period, the target domain is any domain among the multiple domains, and any hyperedge in the local hypergraph and the global hypergraph represents the interaction items of any user in the corresponding time period. An interactive project time series subsequence composed of interactive projects; inputting the local hypergraph and the global hypergraph into a cross-domain sequence recommendation model to predict the user's to-be-interacted projects; wherein the cross-domain sequence recommendation model is used to extract the collaboration information between the interactive project time series subsequences of different users and within the interactive project time series subsequences based on the local hypergraph and the global hypergraph through a hypergraph convolutional network, obtain the project features of the interactive projects containing the collaboration information, determine the user-project interaction features of any user based on the project features corresponding to the interactive projects in the interactive project time series subsequence in multiple time periods, and predict the to-be-interacted projects based on the user-project interaction features.

[0023] It can be seen that this specification constructs a local hypergraph of a single domain and a global hypergraph containing multiple domains based on the time series of historical user interaction items corresponding to multiple users. The hyperedge represents the time series subsequence of interaction items composed of the interaction items of any user in the corresponding time period. The collaborative information between the time series subsequences of interaction items of different users and within the time series subsequences of interaction items are extracted through the hypergraph convolutional network. The influence relationship between the interaction items of the same user in a certain period and the influence relationship between the interaction items of different users in a certain period are taken into account, so as to obtain richer information to predict the items to be interacted, and then make item recommendations, which can improve the accuracy of item recommendations and thus improve user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a recommendation system architecture diagram of a cross-domain sequence recommendation method based on a hypergraph convolutional neural network provided in this specification;

[0025] Figure 2 It is a flowchart of a cross-domain sequence recommendation method based on a hypergraph convolutional neural network provided in this specification;

[0026] Figure 3 This is a schematic diagram of the structure of a cross-domain sequence recommendation model provided in this specification;

[0027] Figure 4 It is a schematic diagram of the structure of a multi-attention transfer gating module provided in this manual;

[0028] Figure 5 This is a flowchart of a cross-domain sequence recommendation model training method provided in this specification;

[0029] Figure 6 It is a structural schematic diagram of a cross-domain sequence recommendation device based on a hypergraph convolutional neural network provided in this specification;

[0030] Figure 7 It is a structural block diagram of a computing device provided in this specification. DETAILED DESCRIPTION

[0031] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

[0032] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0033] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0034] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0035] First, the terms involved in one or more embodiments of this specification are explained.

[0036] Hypergraph is an advanced data structure used to represent complex relationships and multi-dimensional connectivity. Unlike traditional graphs, the edges of a hypergraph can connect more than two nodes, called hyperedges. This enables hypergraphs to represent one-to-many or many-to-many relationships between multiple entities, making them ideal for complex scenarios that need to capture high-dimensional relationships and complex interactions.

[0037] The degree matrix is ​​an important concept in graph theory, mainly used to describe the degree of each vertex in the graph. The degree matrix is ​​a diagonal matrix, and the elements on its diagonal represent the degree of each vertex in the graph. In an undirected graph, the degree of a vertex refers to the number of edges directly connected to the vertex. In a directed graph, the degree of a vertex is divided into out-degree and in-degree, which respectively represent the number of edges starting from the vertex and the number of edges pointing to the vertex.

[0038] Graph convolution is a deep learning technique for graph structured data. Its core idea is to use the node's neighbor information to update the node's feature representation while retaining the global structure information of the graph.

[0039] In this specification, a cross-domain sequence recommendation method based on a hypergraph convolutional neural network is provided. This specification also involves a cross-domain sequence recommendation model training method, a cross-domain sequence recommendation device based on a hypergraph convolutional neural network, a recommendation system, a computing device, and a computer-readable storage medium, which are described in detail one by one in the following embodiments.

[0040] See also Figure 1 , Figure 1 This is a recommendation system architecture diagram of a cross-domain sequence recommendation method based on a hypergraph convolutional neural network provided by an embodiment of this specification. It can include client devices and servers. Among them, the client device can be a terminal device such as a mobile phone, a tablet computer, a desktop computer, etc., and the server includes but is not limited to a physical server and a cloud server.

[0041] The server is used to obtain a time sequence of historical user interaction items in multiple domains, wherein the time sequence of historical user interaction items includes interaction items corresponding to multiple users; construct local hypergraphs and global hypergraphs corresponding to different time periods, wherein the local hypergraph is a hypergraph constructed based on interaction items in the time sequence of historical user interaction items in a target domain within the corresponding time period, and the global hypergraph is a hypergraph constructed based on interaction items in the time sequence of historical user interaction items in the multiple domains within the corresponding time period, wherein the target domain is any domain among the multiple domains, and any hyperedge in the local hypergraph and the global hypergraph represents an interaction item time sequence subsequence constituted by interaction items of any user within the corresponding time period; and the local hypergraph is used to generate a hypergraph based on interaction items in the time sequence of historical user interaction items in the target domain within the corresponding time period. The local hypergraph and the global hypergraph are input into a cross-domain sequence recommendation model to predict the user's to-be-interacted items; and the recommendation information corresponding to the to-be-interacted items is sent to the client device corresponding to the user. The cross-domain sequence recommendation model is used to extract the collaboration information between the interaction item time subsequences of different users and within the interaction item time subsequences based on the local hypergraph and the global hypergraph through a hypergraph convolutional network, obtain the item features of the interaction items containing the collaboration information, determine the user-item interaction features of the user based on the item features corresponding to the interaction items in the interaction item time subsequences of any user in multiple time periods, and predict the to-be-interacted items based on the user-item interaction features.

[0042] The client device is used to receive the recommendation information and display the items to be interacted with through a user interaction interface.

[0043] In addition, the historical user interaction item time sequence may be a sequence of interaction items sent from the client device to the server.

[0044] Taking the application scenario of novel recommendation as an example, multiple domains include novel domain and movie domain, obtain the historical user interaction project time series sequence of the novel domain and the movie domain, the time interval corresponding to the historical user interaction project time series sequence is 5 days, and the period is 1 day; construct the local hypergraph and global hypergraph corresponding to the 5 periods respectively, for each period, construct the local hypergraph corresponding to the novel domain, the local hypergraph includes the novels clicked by the user in the period, each user corresponds to a hyperedge, including the effect clicked by the user in the period, and similarly construct the local hypergraph corresponding to the movie domain, and the global hypergraph corresponding to the novel domain and the movie domain, the global hypergraph includes the novels and movies clicked by the user in the period. Any hyperedge includes the novels and movies clicked by the user corresponding to the hyperedge in the period. Input the local hypergraph and the global hypergraph into the cross-domain sequence recommendation model to predict the novels to be interacted by the user; send the recommendation information corresponding to the novel to be interacted to the client device corresponding to the user, and the client device displays the novel through the user interaction interface. In addition, the movie to be interacted by the user can also be predicted, and the movie to be interacted can be pushed to the client device.

[0045] The client device and the server are connected via a network. The network provides a medium for the communication link between the client device and the server. The network may include various connection types, such as wired or wireless communication links or optical fiber cables. The data transmitted by the client device may need to be encoded, transcoded, compressed, etc. before being released to the server.

[0046] The server may include servers that provide various services, such as servers that provide communication services to multiple clients, and servers that process data sent by clients. It should be noted that the server can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. The server can also be a server of a distributed system, or a server combined with a blockchain. The server can also be a cloud server for basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN, Content Delivery Network), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0047] See also Figure 2 , Figure 2 A flowchart of a cross-domain sequence recommendation method based on a hypergraph convolutional neural network provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0048] Step 202: Acquire a time sequence of historical user interaction items of multiple domains, wherein the time sequence of historical user interaction items includes interaction items corresponding to multiple users.

[0049] Among them, domain can be understood as a field, such as a book domain, a movie domain, an e-commerce domain, etc. Multiple can be understood as greater than or equal to two. Interactive items can be understood as items with user interactive behaviors, such as books clicked by users in the book domain, movies clicked in the movie domain, and novels clicked in the novel domain. Users can click on different items in each app. The time series sequence of historical user interaction items can be understood as a time series sequence composed of interactive items corresponding to multiple users in any domain of the multiple domains. In the time series sequence, the interactive items are arranged in chronological order, and different domains correspond to different time series sequences of historical user interaction items.

[0050] Step 204: construct a local hypergraph and a global hypergraph corresponding to different time periods, wherein the local hypergraph is a hypergraph constructed based on the interaction items in the historical user interaction item time series sequence of the target domain within the corresponding time period, and the global hypergraph is a hypergraph constructed based on the interaction items in the historical user interaction item time series sequence of the multiple domains within the corresponding time period, the target domain is any domain among the multiple domains, and any hyperedge in the local hypergraph and the global hypergraph represents an interaction item time series subsequence constituted by the interaction items of any user within the corresponding time period.

[0051] Among them, the time period can be understood as the time period divided according to the preset interval within the time interval corresponding to the historical user interaction project time series. The preset time interval can be set according to the actual scenario, such as 1 day, 1 hour, 0.5 hour, 10 minutes, etc. For example, the time interval corresponding to the historical user interaction project time series is the past 1 day, with 2 hours as the preset time interval, it can be divided into 6 time periods, namely T1, T2, T3, T4, T5 and T6, each time period corresponds to a different local hypergraph and global hypergraph.

[0052] A local hypergraph can be understood as a hypergraph constructed based on the interaction items in the temporal sequence of historical user interaction items in a single domain within any time period. For example, the local hypergraph corresponding to the T1 period in the movie domain is constructed based on the movie items in the temporal sequence of historical user interaction items in the movie domain within the T1 period. For example, the T1 period includes movie A clicked by user 1 and user 3, movie B clicked by user 2 and user 3, movie C clicked by user 4 and user 5, and movie D clicked by user 1 and user 5. The local hypergraph is constructed based on movie A, movie B, movie C, and movie D.

[0053] The global hypergraph can be understood as a hypergraph constructed based on the interaction items in the temporal sequence of historical user interaction items in multiple domains within any time period. For example, the global hypergraph corresponding to the T1 period of the movie domain and the book domain is constructed based on the movie items in the temporal sequence of historical user interaction items in the movie domain within the T1 period and the book items in the temporal sequence of historical user interaction items in the book domain within the T1 period.

[0054] Both the local hypergraph and the global hypergraph include a vertex set, a hypergraph edge set, an association matrix, a degree matrix of vertices and hyperedges, and a time period corresponding to the hypergraph. The vertex set includes the interactive items in the time period. The vertex set refers to all the supernodes in this time period. The supernodes are interactive items. The hypergraph edge set includes the hyperedges corresponding to the users in the time period. The items that a user has interacted with in this time period can be represented as multiple supernodes. These supernodes are connected to form a hyperedge. The association matrix represents the ownership relationship between the corresponding vertices and hyperedges. The degree matrix of the vertex is a diagonal matrix. The elements on the diagonal are the degrees of each vertex. The degree of the vertex represents the number of edges associated with the vertex. The degree matrix of the hyperedge is a diagonal matrix. The elements on the diagonal are the degrees of each hyperedge. The degree of the hyperedge represents the number of vertices associated with the hyperedge.

[0055] The interactive project time series subsequence can be understood as the interactive project time series subsequence of any user's interactive project time series sequence within the time interval corresponding to the historical user interactive project time series sequence. For example, in 6 time periods, any user's interactive project time series sequence can also be divided into 6 subsequences. The interactive project time series sequence of any user includes the interactive projects of the user. The projects that a user has interacted with in a certain period of time constitute the interactive project time series subsequence. An entire user sequence is divided into many subsequences according to the time period. For example, a user has a sequence of project 1, project 2, project 3, project 4, project 5, project 6, project 7, project 8, project 9, which is divided into three sections according to time: project 1, project 2, project 3, project 4, project 5, project 6, project 7, project 8, project 9. Each section is a subsequence, and each hyperedge corresponds to a unique subsequence.

[0056] It is understandable that in this specification, the local hypergraphs of the same domain in different time periods constitute a local hypergraph flow. Based on the local hypergraph flow, the dynamic preferences in the time series of historical user interaction items in a single domain can be captured. First, it is necessary to complete the construction of the dynamic hypergraph and convert the time series of historical user interaction items in a single domain into a series of dynamic hypergraphs. Each hypergraph corresponds to a specific time interval and contains the interaction items of all users in that time period. The hypergraph corresponding to a time period in a single domain is a local hypergraph. Formally, taking domain A as an example, the time series of historical user interaction items is represented as S A =[v1,v2,...,v i ,...], where v i Represents the user's interactive items in domain A. This specification divides the entire historical user interactive item time series into several time periods T1 = [0, t1], T2 = [t1, t2], ..., T n =[t n-1 ,t n ], will be in each period T iThe local hypergraph corresponding to the inner construction is defined as:

[0057]

[0058] Where A represents the domain A, and i represents the time period. The hypergraph consists of a set of vertices Represents all interactive items in this period, the hypergraph edge set It represents the set of hyperedges connected by multiple supernodes (i.e., interaction items) during the period. Each hyperedge represents a unique interaction subsequence, and one user corresponds to one hyperedge. Represents the association matrix of the hypergraph. When its value is 1, it means that the corresponding node belongs to the corresponding hyperedge. and Denote the diagonal matrices of vertices and hyperedges, respectively, T i Indicates the time period corresponding to the hypergraph. The association matrix represents the relationship between interactive items and hyperedges in the hypergraph. It is a two-dimensional matrix with only 0 and 1 elements. Assuming there are n users (n hyperedges) and m items (m hypernodes), the association matrix is ​​m*n dimensional. When its value is 1, it means that the corresponding node belongs to the corresponding hyperedge, indicating that the user has clicked on this item.

[0059] As time goes by, the topology of these dynamic hypergraphs changes continuously, eventually forming local hypergraph flows:

[0060]

[0061] A represents domain A (same for domain B), and 1-n represents the time period.

[0062] In this specification, the global hypergraph flow can capture cross-domain sequence patterns. Different from the local hypergraph flow, the global hypergraph flow models the cross-domain user interaction sequence in a shared feature space. Specifically, the historical user interaction item time series of domain A and domain B are merged into a cross-domain sequence S M ={B1,A1,B2,...,A i ,B j ,...}, and in each period T i The corresponding global dynamic hypergraph flow is constructed in the network. The construction of the global hypergraph can refer to the local hypergraph, but its goal is to model cross-domain sequential patterns.

[0063]

[0064] Wherein, S represents the cross-domain sequence of the user in domain A and domain B, and 1-n represents the time period.

[0065] Step 206: Input the local hypergraph and the global hypergraph into a cross-domain sequence recommendation model to predict the user's items to be interacted; wherein the cross-domain sequence recommendation model is used to extract the collaboration information between the interaction item time subsequences of different users and within the interaction item time subsequences based on the local hypergraph and the global hypergraph through a hypergraph convolutional network, obtain the item features of the interaction items containing the collaboration information, determine the user-item interaction features of any user based on the item features corresponding to the interaction items in the interaction item time subsequences of multiple time periods, and predict the items to be interacted based on the user-item interaction features.

[0066] Among them, the cross-domain sequence recommendation model can be understood as a model obtained by training the neural network model using a large number of user interaction project time series sequence training samples. The item to be interacted can be understood as the predicted item that the user wants to click next, and the hypergraph convolutional network can be understood as a graph convolutional network that processes hypergraph data. Collaborative information can be understood as the mutual influence relationship information between interactive items of different users and the mutual influence relationship information between interactive items of the same user. Project features can be understood as features of interactive items containing collaborative information, which can be embedded vectors. User-project interaction features can be understood as project interaction features that characterize the user in the time interval corresponding to the historical user interaction project time series sequence, and project interaction features can be understood as features containing user-project interaction interests. Multiple time periods can be understood as time periods within the time interval corresponding to the historical user interaction project time series sequence.

[0067] In an optional implementation of the present specification, the hypergraph convolutional network includes different convolutional structures corresponding to different time periods, and different local hypergraphs and global hypergraphs in any time period correspond to different convolutional structures, respectively. The hypergraph convolutional network extracts the collaborative information between the interactive project time subsequences of different users and within the interactive project time subsequences based on the local hypergraph and the global hypergraph to obtain the project features of the interactive projects containing the collaborative information, which may include: determining the input data of the convolutional structure, wherein the input data includes the project features of the interactive projects in the local hypergraph or the global hypergraph corresponding to the convolutional structure, and the convolution structure corresponding to the global hypergraph. Any item feature in the input data of the convolution structure is a static embedded representation of the interactive item, and any item feature in the input data of the convolution structure corresponding to the local hypergraph is a feature determined based on the item feature output by the convolution structure in the previous time period. If the interactive item appears for the first time in the historical user interactive item time series sequence, then the item feature of the interactive item in the input data is a static embedded representation of the interactive item; the input data is input into the convolution structure, and the collaborative information between the interactive item time series subsequences of different users and within the interactive item time series subsequences is extracted to obtain the item feature of the interactive item output by the convolution structure containing the collaborative information.

[0068] Among them, the convolution structure can be understood as a structure including multiple hypergraph convolution layers, and the project features can be understood as features corresponding to interactive projects. The input data can also include project features of interactive projects in other local hypergraphs or global hypergraphs processed by the convolution structure in the previous period. The convolution structure of the local hypergraph performs convolution processing on the project features of the interactive projects in the local hypergraph or global hypergraph corresponding to itself, and does not perform convolution processing on the project features of interactive projects in other local hypergraphs or global hypergraphs, and transmits them to the convolution structure of the subsequent period. The convolution processing is a graph convolution processing process, that is, the collaborative information between the interactive project time subsequences of different users and within the interactive project time subsequences is extracted to obtain the project features of the interactive projects output by the convolution structure containing the collaborative information. After being processed by the convolution structure, the project features can contain richer information and increase the project features. In addition, the convolution structure corresponding to the local hypergraph transmits the project features of the interactive projects corresponding to other local hypergraphs to the convolution structure corresponding to the local hypergraph of the same domain in the next period. Any item feature in the input data of the convolution structure corresponding to the local hypergraph is a feature determined based on the item features output by the convolution structure corresponding to the local hypergraph of the same domain in the previous time period and the item features output by the convolution structure corresponding to the global hypergraph.

[0069] The static embedding representation can be understood as the initial embedding vector obtained by vectorizing the interactive items. The initial embedding vector can be a randomly initialized vector. Each interactive item has a corresponding id (identifier), and a static embedding representation of the item can be generated and stored in a table. Each id corresponds to a static embedding representation, and when a static embedding representation is needed, it is retrieved according to the id index. For the global hypergraph, the convolution structure of each time period takes the corresponding static embedding representation from the table according to the interactive items contained in the global hypergraph. For the local hypergraph, the local hypergraph of the first time period can take the corresponding static embedding representation according to the interactive items contained in the local hypergraph. The local hypergraph of the subsequent time periods and the interactive items that appear for the first time can also take the static embedding representation from the table.

[0070] In this specification, through different convolution structures corresponding to different time periods, the information extracted from the local hypergraph and the global hypergraph in the short term is propagated backward until the project features of the complete time interval are obtained, which can capture the user's dynamic preferences and thus improve the accuracy of project prediction.

[0071] In this specification, in order to explore the collaborative relationship between items and sequences in a local hypergraph, a hypergraph convolutional network is designed to propagate information in the hypergraph through a message passing mechanism:

[0072]

[0073] Among them, A represents the domain A, i represents the time period, and l represents the lth convolutional layer. Indicates T i The node representation (i.e., item feature) of the l+1th layer of the hypergraph convolutional layer of the local hypergraph of domain A in the time period. and Denote the diagonal matrices of vertices and hyperedges respectively, Represents the incidence matrix of the hypergraph. represents the hyperedge weight matrix, assigning all its weights to 1, represents the trainable weight matrix associated with the lth layer, and δ is the activation function.

[0074] In this specification, the representation of the project can be initialized as a static embedding representation, and the project representation (i.e., project features) can be dynamically updated in each period. By stacking multiple convolutional layers, high-order connections of the hypergraph topology can be extracted to clarify the collaborative relationship within and between sequences. i After L layers of convolution, the node embedding in the local hypergraph of domain A is expressed as Node embeddings represent item features corresponding to interacting items.

[0075] In the global hypergraph, the same information propagation method as the local hypergraph is adopted to extract the collaborative relationship between cross-domain projects and sequences through the hypergraph convolutional network. i After L layers of convolution, the node embedding in the global hypergraph is expressed as

[0076] In an optional implementation of the present specification, the cross-domain sequence recommendation model also includes a gating module connected to the output end of the convolution structure corresponding to the local hypergraph and the global hypergraph of the same time period, and different time periods correspond to different gating modules, wherein the gating module is used to: for any interactive project, determine internal features and external features based on the project features output by the convolution structure corresponding to the target local hypergraph and the target global hypergraph, and fuse the internal features and the external features to obtain the gated output features; wherein the gated output features are used to fuse with the project features output by the convolution structure corresponding to the target local hypergraph to obtain the project features corresponding to the interactive project, so that the next convolution structure corresponding to the local hypergraph where the interactive project exists is used for the interactive project. The project features are convolved, and the target local hypergraph and the target global hypergraph are the local hypergraph and the global hypergraph corresponding to the same time period as the gating module; and the internal features are the features obtained by fusing the project features of the interactive project output by the convolution structure corresponding to the target local hypergraph and the target global hypergraph, the external features are the features obtained by fusing the external node features and the external hyperedge features, the external node features are the features obtained by fusing the project features of the cross-domain projects, the cross-domain projects are cross-domain interactive projects selected based on the similarity with the interactive projects, the external hyperedge features are the features obtained based on the features of the hyperedge where the interactive project is located, and the hyperedge features are the features determined based on the project features of the interactive projects on the hyperedge.

[0077] Among them, the gating module can be understood as a module for cross-domain information transmission, which fuses the project features corresponding to the local hypergraph with the project features corresponding to the global hypergraph according to the preset logic, and transmits them to the convolution structure corresponding to the local hypergraph in the subsequent period. The output connection of the convolution structure corresponding to the local hypergraph and the global hypergraph in the same period can be understood as the project features output by the convolution structure corresponding to the local hypergraph and the global hypergraph input into the gating module of the same period. Specifically, the gated output features are used to fuse the project features of any interactive project output by the convolution structure corresponding to the target local hypergraph to obtain the project features corresponding to the interactive project. Fusion can be understood as the process of merging or integrating different features to generate a more comprehensive and robust feature representation, thereby improving the performance and accuracy of the model. Internal features can be understood as features corresponding to the same interactive project, and external features can be understood as features associated with other projects. In addition to the inherent attributes of the project (internal features), the project will also express some sequence features of projects related to the sequence, that is, external features. External features can be understood as a project may show different features in different sequences, or the weights of various features of a project in different sequences are different. The cross-domain project can be understood as an interactive project in a different domain from any of the interactive projects.

[0078] In this specification, the project features of the convolutional structures corresponding to the local hypergraph and the global hypergraph of the same project are fused to enhance the intra-domain sequence learning. The correlation with other cross-domain projects and the correlation of different user preferences are considered, and external feature alignment is performed to ensure that in the process of cross-domain information transmission, useful information is retained to the maximum extent and irrelevant or harmful interference is reduced, thereby improving the recommendation effect.

[0079] In an optional implementation of the present specification, determining internal features may include: determining the attention scores corresponding to the target local hypergraph and the target global hypergraph based on the project features of any interactive project output by the convolution structures corresponding to the target local hypergraph and the target global hypergraph; calculating the retention ratio corresponding to the target local hypergraph and the transfer ratio corresponding to the target global hypergraph based on the attention scores; and fusing the project features of any interactive project output by the convolution structures corresponding to the target local hypergraph and the target global hypergraph based on the retention ratio and the transfer ratio to obtain internal features.

[0080] Among them, the attention score can be understood as an intermediate value before calculating the attention weight, which reflects the degree of attention paid by the model to different data. The retention ratio can be understood as the weight of retaining the project features corresponding to the local hypergraph, and the transfer ratio can be understood as the weight of transferring the project features corresponding to the global hypergraph backward.

[0081] In this specification, the weights of the corresponding project features of the local hypergraph and the global hypergraph are determined based on the attention mechanism during fusion, and then fused, which can ensure that the fused internal features retain more useful information and improve the recommendation effect.

[0082] In an optional implementation of the present specification, determining external node features may include: calculating a first similarity between a project feature corresponding to any interactive project in the target global hypergraph and a project feature of a cross-domain interactive project in the target global hypergraph; screening a preset number or a preset proportion of cross-domain interactive projects with the highest first similarity to obtain a cross-domain project set; and aggregating project features of the cross-domain project set to obtain external node features.

[0083] Among them, the first similarity can characterize the similarity between the project features of any interactive project and the project features of interactive projects in the target global hypergraph that do not belong to the same domain as any interactive project features. The vector similarity between project features can be calculated, for example, cosine similarity, etc., to calculate the first similarity. It can be understood that the project features selected for calculating the similarity can be the output of the convolution structure corresponding to the global hypergraph. This specification can take a preset number of cross-domain interactive projects or a preset proportion of cross-domain interactive projects, perform project feature aggregation, and obtain external node features.

[0084] In this specification, the project features of the cross-domain interactive projects are screened out based on similarity and integrated, which can ensure the stability of cross-domain information transmission and avoid introducing interference information.

[0085] In an optional implementation of the present specification, determining the external hyperedge feature may include: determining the hyperedge where any interactive item in the target local hypergraph is located to obtain a first target hyperedge, and aggregating features of the first target hyperedge to obtain a local hyperedge feature; determining the hyperedge where any interactive item in the target global hypergraph is located to obtain a second target hyperedge, and aggregating features of the second target hyperedge to obtain a global hyperedge feature; and fusing the local hyperedge feature and the global hyperedge feature to obtain the external hyperedge feature of any interactive item.

[0086] It can be understood that this specification constructs a local hypergraph and a global hypergraph based on a historical user interaction item time series sequence including interaction items corresponding to multiple users. In the local hypergraph and the global hypergraph, any interaction item can exist on multiple hyperedges.

[0087] In this specification, for any interactive item, by integrating the features of the hyperedge where it is located, it is possible to obtain the influence information between different households with similar interests, thereby improving the recommendation effect.

[0088] The gating module realizes stable information transfer and integration between local and global hypergraph flows. It is a key component connecting local and global hypergraph flows. The design of the gating module takes into account the dynamics of user preferences and the complexity of cross-domain information, aiming to avoid negative transfer and improve the performance of cross-domain recommendations. Fine-grained knowledge transfer can be achieved in the following ways:

[0089] Internal feature fusion: The goal is to integrate intrinsic item properties that are represented differently in local and global spaces to enhance intra-domain sequence learning. This is achieved by calculating attention scores that determine the proportion of local information to be preserved and the proportion of global information to be transferred during the fusion process.

[0090] First, the attention score between the local and global candidate sets is calculated.

[0091]

[0092] Among them, e b,x and e s,x Respectively represent the node embedding representation of the interaction item x as a hypergraph node in the local hypergraph and the global hypergraph, and W I are all trainable weight parameters, and σ is the tanh activation function. l and g to denote the attention scores of node embedding representations in local and global hypergraphs, and [] denotes a concatenation operation.

[0093] According to the attention scores, the node embedding representations in the local and global hypergraph streams are fused.

[0094] e ii,x =re b,x +(1-r)e s,x ;

[0095]

[0096] Among them, r is the proportion of local information retention, and 1-r is the proportion of global information transmission. ii,x is the internal characteristics of the obtained item x. exp(x) represents the natural exponential function.

[0097] External feature alignment aims to explore the association between externalized features of common items in different sequence contexts and reduce domain bias. This process involves two agents: the external node agent (i.e., the filtered interactive items) and the external hyperedge agent (i.e., the filtered hyperedges), which are used for item-level transfer and user-level transfer respectively. External features can be understood as features extracted and aggregated from the two agents, and alignment can be understood as an operation that processes data with different data distributions or different feature spaces, thereby eliminating the differences between data to a certain extent.

[0098] External node agents: Node agents are selected by measuring the implicit associations between items in the global dynamic hypergraph, and then the node agents are aggregated using an attention mechanism.

[0099]

[0100] Among them, W N is a learnable parameter, e np,x It represents the fusion state of the node agent, that is, the external node features corresponding to the interactive items. The node proxy set selected to measure the implicit association between items in the global dynamic hypergraph is the cross-domain item set. s,x represents the node embedding representation of the interaction item x in the global hypergraph as a hypergraph node, e s,i represents the node embedding representation of the interactive item i in the node agent collection as a hypergraph node in the global hypergraph, that is, the external node feature. a(x,i) represents the weight designed according to the attention mechanism, the purpose is to better extract information, e s,j The node embedding representation of the interactive items in the node proxy collection is represented. This specification traverses each node in the node proxy collection. For each item x in domain B in the global candidate set, based on the similarity of super nodes, the top K most similar nodes are selected from domain A in the global as the node proxy collection.

[0101] The external hyperedge proxy obtains the representation of the hyperedge by extracting the dynamic representation after the hypergraph convolution, and then aggregates the local and global hyperedge proxies to generate an integrated sequence representation for external features.

[0102] First, we obtain the dynamic representation of all hyperedges by aggregating the nodes on all hyperedges:

[0103]

[0104] The resulting matrix Representing a local hypergraph The superscript 2 indicates the second period. In this embodiment, the local hypergraph As an example, the representation of hyperedges in other local hypergraphs and global hypergraphs can refer to this example. For each candidate node x in the local hypergraph, take out the node to which x belongs. All hyperedges in , as local hyperedge proxies

[0105]

[0106] Then, as in the above formula, a layer of average pooling is used to obtain the fusion state u of the hyperedge agent of the local hypergraph l,x ,ub,t Represents a local hyperedge agent. The same method can be used to obtain the fusion state u of the hyperedge agent on the global hypergraph. g,x .

[0107] e ep,x =ru l,x +(1-r)u g,x ;

[0108] Among them, r is the proportion of local information retention, and 1-r is the proportion of global information transmission. ep,x is an external hyperedge feature generated by combining the local and global fusion states of the hyperedge agent (i.e., local hyperedge features and global hyperedge features).

[0109] Furthermore, the fused internal features, external node features, and external hyperedge features are used to generate dynamic item embeddings, which are then fused with the output of the convolutional structure corresponding to the local hypergraph of the same period as the input of the convolutional structure of the local hypergraph of the next period. In this way, the stable transmission and integration of cross-domain information is achieved through the dynamic hypergraph flow.

[0110] In an optional implementation of the present specification, determining the user-item interaction features of any user based on the item features corresponding to the interaction items in the interaction item time series subsequences of multiple time periods may include: combining the interaction item time series subsequences of any user in multiple time periods into a user interaction item time series sequence in a target domain; fusing the item features output by the convolutional structure corresponding to the local hypergraph of the interaction items in the user interaction item time series sequence in a first preset time period into the user-item interaction features of the user based on a fusion weight; wherein the fusion weight is determined based on the item features of the specified interaction items in the user interaction item time series sequence.

[0111] Among them, the target domain can be understood as any domain among multiple domains, and this specification can predict the items to be interacted with by the user in the target domain. That is, what is fused is the project features output by the convolution structure corresponding to the local hypergraph of the target domain, and the fusion weight can be understood as the weight corresponding to the project features when multiple project features are fused. The preset time period can be understood as a specified time period. For example, the first preset time period is the time period when the user last interacted with the interactive item within the time interval corresponding to the historical user interactive item time series sequence, that is, it can be specified that the first preset time period corresponding to any interactive item is the time period when the user last clicked on the item. Different interactive items can correspond to different time periods.

[0112] In this specification, by presetting time periods, you can select time periods close to predicted items to be interacted with for project feature fusion, and specify interactive items or specify items close to predicted items to be interacted with for weight calculation. The closer to the item to be interacted with, the higher the correlation with the item to be interacted with. It also combines the user's short-term and long-term interests to improve the accuracy of the prediction of the item to be interacted with.

[0113] In this specification, a sequence representation containing rich information is generated, combining the user's short-term and long-term interests, forming a complete single-domain sequence by recombining subsequences, and using a self-attention layer to extract long-term interest patterns.

[0114]

[0115] x a,u represents the final representation of the sequence Su in domain A, i.e., the user-item interaction feature. Su represents the temporal sequence of user interaction items of user u, and W Q , W K and W v is a learnable training parameter, x a,m Represents the node embedding representation of the time period corresponding to the last item in the time series of user interaction items. This time period can be the last time the user clicks on the item. a,i The node embedding representation of the time period corresponding to a certain interactive item in the sequence Su in the domain A can be the last time period when the user clicks on the item. D is the embedding dimension, and the scaling factor This is to avoid the inner product value being too large, especially when the dimension is high.

[0116] In an optional implementation of the present specification, predicting the item to be interacted based on the user-item interaction feature may include: calculating a second similarity between the user-item interaction feature and the item feature output by the convolution structure corresponding to the local hypergraph of the interaction item in a second preset time period; and predicting the item to be interacted based on the second similarity.

[0117] The second similarity may represent the degree of similarity between project features, and may be a vector similarity between project features. The second similarity between the user-project interaction feature and the project feature output by the convolution structure corresponding to the local hypergraph of the interactive project of the target domain in the second preset time period may be calculated. Furthermore, in an optional implementation, the interactive project with the highest second similarity may be determined as the project to be interacted.

[0118] Specifically, the second preset time period can be the time period when the interactive item appears for the last time within the time interval corresponding to the historical user interaction item time series sequence. The accuracy of item prediction can be improved by outputting the item features of the most recent time period. For example, the time period to be predicted is T15, item 1 last occurred at T13, and item 2 last occurred at T14. This manual uses the item features of item 1 output by convolution at T13 to calculate the similarity with the user-item interaction features, and uses the item features of item 2 output by convolution at T14 to calculate the similarity with the user-item interaction features. The closer the time periods, the closer the item features of the items.

[0119] In this specification, based on the similarity between the user-item interaction features and the item features of the interaction items, the accuracy of predicting the items to be interacted with can be improved.

[0120] It can be seen that this specification constructs a local hypergraph of a single domain and a global hypergraph containing multiple domains based on the time series of historical user interaction items corresponding to multiple users. The hyperedge represents the time series subsequence of interaction items composed of the interaction items of any user in the corresponding time period. The collaborative information between the time series subsequences of interaction items of different users and within the time series subsequences of interaction items are extracted through the hypergraph convolutional network. The influence relationship between the interaction items of the same user in a certain period and the influence relationship between the interaction items of different users in a certain period are taken into account, so as to obtain richer information to predict the items to be interacted, and then make item recommendations, which can improve the accuracy of item recommendations and thus improve user experience.

[0121] This specification extracts the complex collaborative relationship of the sequence within the domain, while considering the internal collaboration (collaboration within the same sequence) and external collaboration (collaboration between different sequences) of user behavior. In the same sequence, the items that users interact with within a certain period often have similar characteristics. Therefore, even if the conversion within the sequence is not continuous, it will cooperate with each other. Between different sequences, users with common interests are more likely to perform similar behavior patterns hidden under the interaction trajectory. Both aspects of the collaborative relationship may have a profound impact on the representation learning of items and sequences. This specification can make full use of the complex dependencies between sequences. In addition, the capture and stable transmission of inter-domain interaction patterns can be achieved. The conversion of items within the sequence can indicate the evolution of user preferences, which does not always follow a stable pattern. From a fine-grained perspective, users' short-term interests may occasionally fluctuate, and directly transferring each pattern (i.e., useful information) across domains may introduce noise to the two domains, affecting the recommendation effect. At the same time, the factors that determine the interaction are a complex combination of internal and external attributions. In order to effectively utilize cross-domain information, this embodiment distinguishes internal and external features and adopts different transmission strategies accordingly, which helps to maximize the retention of useful information and reduce irrelevant or harmful interference during cross-domain information transmission, which is more in line with the needs of actual scenarios.

[0122] See also Figure 3 , Figure 3 A schematic diagram of the structure of a cross-domain sequence recommendation model provided according to an embodiment of the present specification is shown, wherein the static item embedding of domain A includes the static embedding representation corresponding to all interactive items in the historical user interaction item time sequence of domain A, and a local hypergraph flow of domain A is constructed. Represents the local hypergraph of domain A from time period 1 to n, each local hypergraph corresponds to a multi-layer convolution structure, Including interactive projects: A1, A2, A3, Including interactive items: A1, A2, A3, A4, A5, A5, A2, A4 form a hyperedge, A1, A2, A3 form a hyperedge, Including interactive items: A1, A2, A3, A4, A5, A6, A7, the static item embedding of domain B includes the static embedding representation corresponding to all interactive items in the temporal sequence of historical user interactive items in domain B, and constructs the local hypergraph flow of domain B. represents the local hypergraph of domain B from time 1 to n, B1 to B6 represent the interactive items of domain B, and construct the global hypergraph flow. Indicates the global overclocking of time periods 1 to n Figures 1 to 7 Represents the interactive items, the multi-attention transfer gating represents the aforementioned gating module, and the stream interest generator is used to generate user-item interaction features and then predict the next item. For example, domain A is the movie domain, predict the next movie, domain B is the book domain, predict the next book.

[0123] See also Figure 4 , Figure 4 The schematic diagram of the structure of the multi-attention transfer gating module provided according to the embodiment of the present specification is shown. Taking the multi-attention transfer gating module corresponding to the second period as an example, the global candidate node is project 2, the local candidate node is B2, the global candidate node and the local candidate node are the same project, and internal feature fusion is performed, that is, the project features of the global candidate node and the local candidate node are fused to obtain the internal feature E ii , Includes items 1 to 5, from Select the global hyperedge agent, including the hyperedge composed of project 1, project 1, project 2, project 2 and project 3, and the hyperedge composed of project 4, project 1, project 3 and project 2. Includes items 1 to 4, represented by B1, B2, B3, B4, from Select local hyperedge agents, including the hyperedges formed by project 1 and project 2, and the hyperedges formed by project 4, project 3, and project 2. Take the global hyperedge agent and the local hyperedge agent as the external hyperedge agent, and fuse the features of the external hyperedge agent to obtain the external hyperedge feature E ep, select the external node agent of project 2, transfer the external node agent, and obtain the external node feature E np , perform state fusion, and combine the features after state fusion with The output features of the corresponding convolutional structures are fused to obtain the fused features, and the fused features and the ones selected from the static embedding are combined. Correspondingly, the static embeddings corresponding to the items (B5, B6) that have not appeared in the previous hypergraph represent the input The corresponding convolution structure.

[0124] This manual conducts dual hypergraph flow modeling, and by constructing local and global hypergraph flows, information propagation is performed to fully capture and model the sequence collaboration relationship within and across domains. Through multi-attention transfer gating, information transfer and integration are achieved between local and global hypergraph flows, and internal feature fusion and external feature alignment are further achieved, ensuring the stability and accuracy of cross-domain pattern transfer. In addition, the stream interest generator generates information-rich representations containing users' long-term preference patterns, thereby improving the performance of the recommendation system.

[0125] It can be seen that this specification adopts a cross-domain sequence recommendation method, which can effectively alleviate the data sparsity problem by utilizing shared sequence patterns between different fields, and by utilizing the user's interaction data in multiple fields, it helps the model to more comprehensively understand the user's interest preferences, thereby improving the accuracy of sequence recommendations. In order to effectively mine and integrate cross-domain sequence patterns, avoid introducing noise, and perform stable cross-domain information transmission, a cross-domain sequence recommendation method based on a hypergraph convolutional neural network is provided, which models the sequence collaboration relationship of each domain by constructing a double hypergraph stream for user sequence fragments in each domain in time periods, and designs multi-attention transfer gating to realize internal feature fusion and external feature alignment at the connection of the hypergraph, completing the stable transmission and fusion of cross-domain information, avoiding negative transmission, effectively alleviating the sparsity problem of single-domain sequence data, and generating information-rich sequential representations through a stream interest generator, thereby improving the overall performance of the recommendation system.

[0126] See also Figure 5 , Figure 5 A flowchart of a cross-domain sequence recommendation model training method provided according to an embodiment of this specification is shown, including:

[0127] Step 502: acquiring user interaction item time series samples of multiple domains and label information corresponding to the user interaction item time series samples from a training sample set, wherein the user interaction item time series samples include interaction items corresponding to multiple users.

[0128] The tag information may be the next item clicked by each user. It is understandable that the next item clicked by the user is different in different domains.

[0129] Step 504: construct a local hypergraph and a global hypergraph corresponding to different time periods, wherein the local hypergraph is a hypergraph constructed based on the interaction items in the user interaction item time series sample of the target domain within the corresponding time period, and the global hypergraph is a hypergraph constructed based on the interaction items in the user interaction item time series sample of the multiple domains within the corresponding time period, the target domain is any domain among the multiple domains, and any hyperedge in the local hypergraph and the global hypergraph represents an interaction item time series subsequence constituted by the interaction items of any user within the corresponding time period.

[0130] Step 506: Input the local hypergraph and the global hypergraph into a preset model to predict the user's items to be interacted with, wherein the preset model is used to extract the collaborative information between the interactive item time subsequences of different users and within the interactive item time subsequences based on the local hypergraph and the global hypergraph through a hypergraph convolutional network, obtain the item features of the interactive items containing the collaborative information, determine the user-item interaction features of any user based on the item features corresponding to the interactive items in the interactive item time subsequences of multiple time periods, and predict the items to be interacted based on the user-item interaction features.

[0131] Step 508: Calculate the training loss using the items to be interacted with and the label information.

[0132] Step 510: Update the parameters of the preset model based on the training loss to obtain the preset model with updated parameters.

[0133] If the training end condition is not met, the step of acquiring user interaction item time series samples of multiple domains and label information corresponding to the user interaction item time series samples from the training sample set is continued.

[0134] Step 512: When the training end condition is met, the preset model after the current parameters are updated is determined as the cross-domain sequence recommendation model.

[0135] The training end condition may be that the training loss is less than a preset threshold, or that the number of model iterations reaches a preset threshold, etc.

[0136] It can be seen that during the training process, this specification constructs a local hypergraph of a single domain and a global hypergraph containing multiple domains based on the historical user interaction project time series sequences corresponding to multiple users. The hyperedges represent the interaction project time series subsequences composed of the interaction projects of any user in the corresponding time period. The collaborative information between the interaction project time series subsequences of different users and within the interaction project time series subsequences is extracted through the hypergraph convolutional network. The influence relationship between the interaction projects of the same user in a certain period and the influence relationship between the interaction projects of different users in a certain period are taken into account, and richer information is obtained to predict the items to be interacted. The model learns rich information, which can improve the accuracy of item recommendations, thereby improving user experience.

[0137] Corresponding to the above method embodiment, this specification also provides a cross-domain sequence recommendation device embodiment based on a hypergraph convolutional neural network, Figure 6 FIG. 1 shows a schematic diagram of a cross-domain sequence recommendation device based on a hypergraph convolutional neural network provided by an embodiment of the present specification. Figure 6 As shown, the device comprises:

[0138] A sequence acquisition module 602 is used to acquire a time sequence sequence of historical user interaction items of multiple domains, wherein the time sequence sequence of historical user interaction items includes interaction items corresponding to multiple users;

[0139] A hypergraph construction module 604 is used to construct a local hypergraph and a global hypergraph corresponding to different time periods, wherein the local hypergraph is a hypergraph constructed based on the interaction items in the historical user interaction item time sequence of the target domain in the corresponding time period, and the global hypergraph is a hypergraph constructed based on the interaction items in the historical user interaction item time sequence of the multiple domains in the corresponding time period, the target domain is any domain among the multiple domains, and any hyperedge in the local hypergraph and the global hypergraph represents an interaction item time sequence subsequence composed of the interaction items of any user in the corresponding time period;

[0140] An item prediction module 606 is used to input the local hypergraph and the global hypergraph into a cross-domain sequence recommendation model to predict items to be interacted with by the user;

[0141] Among them, the cross-domain sequence recommendation model is used to extract the collaborative information between the interactive project time subsequences of different users and within the interactive project time subsequences based on the local hypergraph and the global hypergraph through a hypergraph convolutional network, obtain the project features of the interactive projects containing the collaborative information, determine the user-project interaction features of any user based on the project features corresponding to the interactive projects in the interactive project time subsequences of multiple time periods, and predict the project to be interacted based on the user-project interaction features.

[0142] In an optional implementation of the present specification, the hypergraph convolutional network includes different convolutional structures corresponding to different time periods, and different local hypergraphs and global hypergraphs in any time period correspond to different convolutional structures respectively. The item prediction module 606 is specifically used to determine the input data of the convolutional structure, wherein the input data includes item features of interactive items in the local hypergraph or global hypergraph corresponding to the convolutional structure, any item feature in the input data of the convolutional structure corresponding to the global hypergraph is a static embedded representation of the interactive item, and any item feature in the input data of the convolutional structure corresponding to the local hypergraph is a feature determined based on the item features output by the convolutional structure of the previous time period. If the interactive item appears for the first time in the historical user interactive item time series sequence, then the item feature of the interactive item in the input data is a static embedded representation of the interactive item; the input data is input into the convolutional structure, and the collaborative information between the interactive item time series subsequences of different users and within the interactive item time series subsequences is extracted to obtain the item features of the interactive items output by the convolutional structure containing the collaborative information.

[0143] Furthermore, the cross-domain sequence recommendation model also includes a gating module connected to the output end of the convolution structure corresponding to the local hypergraph and the global hypergraph of the same time period, and different time periods correspond to different gating modules, wherein the gating module is used to: for any interactive project, determine the internal features and external features based on the project features output by the convolution structure corresponding to the target local hypergraph and the target global hypergraph, and fuse the internal features and the external features to obtain the gated output features; wherein the gated output features are used to fuse with the project features output by the convolution structure corresponding to the target local hypergraph to obtain the project features corresponding to the interactive project, so that the convolution structure corresponding to the next local hypergraph where the interactive project exists performs convolution processing on the project features, and the target local hypergraph and the target global hypergraph are the local hypergraph and the global hypergraph corresponding to the same time period as the gating module;

[0144] Furthermore, the internal feature is a feature obtained by fusing the project features of the interactive project output by the convolution structure corresponding to the target local hypergraph and the target global hypergraph, the external feature is a feature obtained by fusing the external node feature and the external hyperedge feature, the external node feature is a feature obtained by fusing the project features of the cross-domain project, the cross-domain project is a cross-domain interactive project selected based on the similarity with the interactive project, the external hyperedge feature is a feature obtained based on the feature of the hyperedge where the interactive project is located, and the hyperedge feature is a feature determined based on the project features of the interactive project on the hyperedge.

[0145] Among them, the gate control module includes:

[0146] The internal feature determination submodule is specifically used to determine the attention scores corresponding to the target local hypergraph and the target global hypergraph based on the project features of any interactive project output by the convolution structures corresponding to the target local hypergraph and the target global hypergraph; calculate the retention ratio corresponding to the target local hypergraph and the transfer ratio corresponding to the target global hypergraph based on the attention scores; and fuse the project features of any interactive project output by the convolution structures corresponding to the target local hypergraph and the target global hypergraph based on the retention ratio and the transfer ratio to obtain internal features.

[0147] The external node feature determination submodule is specifically used to calculate the first similarity between the project features corresponding to any interactive project in the target global hypergraph and the project features of the cross-domain interactive project in the target global hypergraph; screen a preset number or a preset proportion of cross-domain interactive projects with the highest first similarity to obtain a cross-domain project set; aggregate the project features of the cross-domain project set to obtain external node features.

[0148] The external hyperedge feature determination submodule is specifically used to: determine the hyperedge where any interactive item in the target local hypergraph is located to obtain a first target hyperedge, aggregate the features of the first target hyperedge to obtain a local hyperedge feature; determine the hyperedge where any interactive item in the target global hypergraph is located to obtain a second target hyperedge, aggregate the features of the second target hyperedge to obtain a global hyperedge feature; and fuse the local hyperedge features and the global hyperedge features to obtain an external hyperedge feature of any interactive item.

[0149] Furthermore, the cross-domain sequence recommendation model also includes a user-item interaction feature determination module, which is specifically used to combine the interaction project time series subsequences of any user in multiple time periods into a user interaction project time series sequence in the target domain; based on the fusion weight, the project features output by the convolutional structure corresponding to the local hypergraph of the interaction projects in the user interaction project time series in the first preset time period are fused into the user-item interaction features of the user; wherein the fusion weight is determined based on the project features of the specified interaction projects in the user interaction project time series.

[0150] Furthermore, the cross-domain sequence recommendation model also includes an item prediction module, which is specifically used to: calculate the second similarity between the user-item interaction features and the item features output by the convolution structure corresponding to the local hypergraph of the interactive items in the second preset time period; and predict the items to be interacted based on the second similarity.

[0151] The above is a schematic scheme of a cross-domain sequence recommendation device based on a hypergraph convolutional neural network of this embodiment. It should be noted that the technical scheme of the cross-domain sequence recommendation device based on a hypergraph convolutional neural network and the technical scheme of the cross-domain sequence recommendation method based on a hypergraph convolutional neural network belong to the same concept. The details of the technical scheme of the cross-domain sequence recommendation device based on a hypergraph convolutional neural network that are not described in detail can be found in the description of the technical scheme of the cross-domain sequence recommendation method based on a hypergraph convolutional neural network.

[0152] It can be seen that this specification constructs a local hypergraph of a single domain and a global hypergraph containing multiple domains based on the time series of historical user interaction items corresponding to multiple users. The hyperedge represents the time series subsequence of interaction items composed of the interaction items of any user in the corresponding time period. The collaborative information between the time series subsequences of interaction items of different users and within the time series subsequences of interaction items are extracted through the hypergraph convolutional network. The influence relationship between the interaction items of the same user in a certain period and the influence relationship between the interaction items of different users in a certain period are taken into account, so as to obtain richer information to predict the items to be interacted, and then make item recommendations, which can improve the accuracy of item recommendations and thus improve user experience.

[0153] Figure 7 The block diagram of a computing device 700 according to an embodiment of the present specification is shown. The components of the computing device 700 include but are not limited to a memory 710 and a processor 720. The processor 720 is connected to the memory 710 via a bus 730, and the database 750 is used to store data.

[0154] The computing device 700 also includes an access device 740 that enables the computing device 700 to communicate via one or more networks 760. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 740 may include one or more of any type of network interface (e.g., a network interface card (NIC)) of wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a world-wide interoperability for microwave access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, and a near field communication (NFC).

[0155] In one embodiment of the present specification, the above components of the computing device 700 and Figure 7 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Figure 7 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0156] The computing device 700 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 700 may also be a mobile or stationary server.

[0157] Among them, the processor 720 is used to execute the following computer program / instructions, which, when executed by the processor, implement the steps of the above-mentioned cross-domain sequence recommendation method based on the hypergraph convolutional neural network, and / or the above-mentioned cross-domain sequence recommendation model training method.

[0158] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the computing device embodiment, since it is basically similar to the cross-domain sequence recommendation method embodiment of the hypergraph convolutional neural network, the description is relatively simple, and the relevant parts can be referred to the partial description of the cross-domain sequence recommendation method embodiment of the hypergraph convolutional neural network.

[0159] An embodiment of the present specification also provides a computer-readable storage medium storing a computer program / instruction, which, when executed by a processor, implements the steps of the cross-domain sequence recommendation method based on a hypergraph convolutional neural network, and / or the cross-domain sequence recommendation model training method.

[0160] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the computer-readable storage medium embodiment, since it is basically similar to the cross-domain sequence recommendation method embodiment of the hypergraph convolutional neural network, the description is relatively simple, and the relevant parts can be referred to the partial description of the cross-domain sequence recommendation method embodiment of the hypergraph convolutional neural network.

[0161] An embodiment of the present specification also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the cross-domain sequence recommendation method of the above-mentioned hypergraph convolutional neural network, and / or the above-mentioned cross-domain sequence recommendation model training method.

[0162] The above is a schematic scheme of a computer program product of this embodiment. It should be noted that the technical scheme of the computer program product and the technical scheme of the cross-domain sequence recommendation method of the hypergraph convolutional neural network mentioned above belong to the same concept, and the details not described in detail in the technical scheme of the computer program product can be found in the description of the technical scheme of the cross-domain sequence recommendation method of the hypergraph convolutional neural network mentioned above.

[0163] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0164] The computer instructions include computer program codes, which may be in source code form, object code form, executable files or some intermediate forms, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0165] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0166] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0167] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that technicians in the relevant technical field can well understand and use this specification. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. A cross-domain sequence recommendation method based on hypergraph convolutional neural network, It is characterized in that include: Acquire a time sequence of historical user interaction items of multiple domains, wherein the time sequence of historical user interaction items includes interaction items corresponding to multiple users; Constructing a local hypergraph and a global hypergraph corresponding to different time periods, wherein the local hypergraph is a hypergraph constructed based on the interaction items in the historical user interaction item time sequence of the target domain in the corresponding time period, and the global hypergraph is a hypergraph constructed based on the interaction items in the historical user interaction item time sequence of the multiple domains in the corresponding time period, the target domain is any domain among the multiple domains, and any hyperedge in the local hypergraph and the global hypergraph represents an interaction item time sequence subsequence constituted by the interaction items of any user in the corresponding time period; Inputting the local hypergraph and the global hypergraph into a cross-domain sequence recommendation model to predict items to be interacted with by the user; Among them, the cross-domain sequence recommendation model is used to extract the collaborative information between the interactive project time subsequences of different users and within the interactive project time subsequences based on the local hypergraph and the global hypergraph through a hypergraph convolutional network, obtain the project features of the interactive projects containing the collaborative information, determine the user-project interaction features of any user based on the project features corresponding to the interactive projects in the interactive project time subsequences of multiple time periods, and predict the project to be interacted based on the user-project interaction features.

2. The method according to claim 1, It is characterized in that The hypergraph convolution network includes different convolution structures corresponding to different time periods, and different local hypergraphs and global hypergraphs in any time period correspond to different convolution structures respectively. The hypergraph convolution network extracts the collaboration information between the interactive project time subsequences of different users and within the interactive project time subsequences based on the local hypergraph and the global hypergraph, and obtains the project features of the interactive project containing the collaboration information, including: Determine input data of the convolution structure, wherein the input data includes item features of interactive items in a local hypergraph or a global hypergraph corresponding to the convolution structure, any item feature in the input data of the convolution structure corresponding to the global hypergraph is a static embedding representation of the interactive item, and any item feature in the input data of the convolution structure corresponding to the local hypergraph is a feature determined based on item features output by the convolution structure in a previous time period, and if the interactive item appears for the first time in the historical user interactive item time series sequence, then the item feature of the interactive item in the input data is a static embedding representation of the interactive item; The input data is input into a convolution structure, and the collaboration information between the interactive project time sequence subsequences of different users and within the interactive project time sequence subsequences is extracted to obtain the project features of the interactive project output by the convolution structure that contain the collaboration information.

3. The method according to claim 2, It is characterized in that The cross-domain sequence recommendation model further includes a gating module connected to the output end of the convolution structure corresponding to the local hypergraph and the global hypergraph of the same time period, and different time periods correspond to different gating modules, wherein the gating module is used to: For any interactive project, internal features and external features are determined based on project features output by the convolutional structure corresponding to the target local hypergraph and the target global hypergraph, and the internal features and the external features are fused to obtain gated output features; The gated output feature is used to fuse with the project feature output by the convolution structure corresponding to the target local hypergraph to obtain the project feature corresponding to the interactive project, so that the convolution structure corresponding to the next local hypergraph where the interactive project exists performs convolution processing on the project feature, and the target local hypergraph and the target global hypergraph are the local hypergraph and the global hypergraph corresponding to the same time period as the gated module; Furthermore, the internal feature is a feature obtained by fusing the project features of the interactive project output by the convolution structure corresponding to the target local hypergraph and the target global hypergraph, the external feature is a feature obtained by fusing the external node feature and the external hyperedge feature, the external node feature is a feature obtained by fusing the project features of the cross-domain project, the cross-domain project is a cross-domain interactive project selected based on the similarity with the interactive project, the external hyperedge feature is a feature obtained based on the feature of the hyperedge where the interactive project is located, and the hyperedge feature is a feature determined based on the project features of the interactive project on the hyperedge.

4. The method according to claim 3, It is characterized in that Determine internal characteristics, including: Determine the attention scores corresponding to the target local hypergraph and the target global hypergraph based on the project features of any interactive project output by the convolution structures corresponding to the target local hypergraph and the target global hypergraph; Calculate the retention ratio corresponding to the target local hypergraph and the transfer ratio corresponding to the target global hypergraph based on the attention score; Based on the retention ratio and the transfer ratio, the project features of any interactive project output by the convolution structure corresponding to the target local hypergraph and the target global hypergraph are fused to obtain internal features.

5. The method according to claim 3, It is characterized in that Determine external node characteristics, including: Calculate a first similarity between a project feature corresponding to any interactive project in the target global hypergraph and a project feature of a cross-domain interactive project in the target global hypergraph; Filtering a preset number or a preset proportion of cross-domain interaction items with the highest first similarity to obtain a cross-domain item set; The project features of the cross-domain project set are aggregated to obtain external node features.

6. The method according to claim 3, It is characterized in that Determine external hyperedge features, including: Determine a hyperedge where any interactive item in the target local hypergraph is located, obtain a first target hyperedge, aggregate features of the first target hyperedge, and obtain a local hyperedge feature; Determine the hyperedge where any interactive item in the target global hypergraph is located, obtain a second target hyperedge, aggregate features of the second target hyperedge, and obtain a global hyperedge feature; The local hyperedge feature and the global hyperedge feature are fused to obtain an external hyperedge feature of any interactive item.

7. The method according to claim 1, It is characterized in that The determining of the user-item interaction feature of any user based on the item features corresponding to the interaction items in the interaction item time sequence subsequences of multiple time periods includes: Combining the interaction item time sequence subsequences of any user in multiple time periods into the user interaction item time sequence sequence of the target domain; Based on the fusion weight, the project features output by the convolution structure corresponding to the local hypergraph of the interactive project in the user interactive project time series in the first preset time period are fused into the user-item interaction features of the user; The fusion weight is determined based on the item features of the specified interactive items in the user interactive item sequence.

8. The method according to claim 1, It is characterized in that The predicting the item to be interacted based on the user-item interaction feature includes: Calculating a second similarity between the user-item interaction feature and an item feature output by a convolutional structure corresponding to a local hypergraph of the interactive item in a second preset time period; The to-be-interacted item is predicted based on the second similarity.

9. A cross-domain sequence recommendation model training method, It is characterized in that include: Acquire user interaction item time series samples of multiple domains and label information corresponding to the user interaction item time series samples from a training sample set, wherein the user interaction item time series samples include interaction items corresponding to multiple users; Constructing local hypergraphs and global hypergraphs corresponding to different time periods, wherein the local hypergraph is a hypergraph constructed based on the interaction items in the user interaction item time series sample of the target domain in the corresponding time period, and the global hypergraph is a hypergraph constructed based on the interaction items in the user interaction item time series sample of the multiple domains in the corresponding time period, the target domain is any domain among the multiple domains, and any hyperedge in the local hypergraph and the global hypergraph represents an interaction item time series subsequence constituted by the interaction items of any user in the corresponding time period; The local hypergraph and the global hypergraph are input into a preset model to predict the user's items to be interacted, wherein the preset model is used to extract the collaboration information between the interactive item time sequence subsequences of different users and within the interactive item time sequence subsequences based on the local hypergraph and the global hypergraph through a hypergraph convolutional network, obtain the item features of the interactive items containing the collaboration information, determine the user-item interaction features of any user based on the item features corresponding to the interactive items in the interactive item time sequence subsequences of multiple time periods, and predict the items to be interacted based on the user-item interaction features; Calculating training loss using the items to be interacted with and label information; The parameters of the preset model are updated based on the training loss to obtain the preset model after parameter update. If the training end condition is not met, the step of obtaining user interaction item time series samples of multiple domains and label information corresponding to the user interaction item time series samples in the training sample set is continued; When the training end condition is met, the preset model after the current parameters are updated is determined as the cross-domain sequence recommendation model.

10. A cross-domain sequence recommendation device based on hypergraph convolutional neural network, It is characterized in that include: A sequence acquisition module, used to acquire a time sequence sequence of historical user interaction items of multiple domains, wherein the time sequence sequence of historical user interaction items includes interaction items corresponding to multiple users; A hypergraph construction module, used to construct local hypergraphs and global hypergraphs corresponding to different time periods, wherein the local hypergraph is a hypergraph constructed based on the interaction items in the historical user interaction item time sequence of the target domain in the corresponding time period, and the global hypergraph is a hypergraph constructed based on the interaction items in the historical user interaction item time sequence of the multiple domains in the corresponding time period, the target domain is any domain among the multiple domains, and any hyperedge in the local hypergraph and the global hypergraph represents an interaction item time sequence subsequence composed of the interaction items of any user in the corresponding time period; An item prediction module, used to input the local hypergraph and the global hypergraph into a cross-domain sequence recommendation model to predict items to be interacted with by the user; Among them, the cross-domain sequence recommendation model is used to extract the collaborative information between the interactive project time subsequences of different users and within the interactive project time subsequences based on the local hypergraph and the global hypergraph through a hypergraph convolutional network, obtain the project features of the interactive projects containing the collaborative information, determine the user-project interaction features of any user based on the project features corresponding to the interactive projects in the interactive project time subsequences of multiple time periods, and predict the project to be interacted based on the user-project interaction features.

11. A recommendation system, It is characterized in that Includes server and client devices, where The server is used to obtain a time sequence of historical user interaction items in multiple domains, wherein the time sequence of historical user interaction items includes interaction items corresponding to multiple users; construct local hypergraphs and global hypergraphs corresponding to different time periods, wherein the local hypergraph is a hypergraph constructed based on interaction items in the time sequence of historical user interaction items in a target domain within the corresponding time period, and the global hypergraph is a hypergraph constructed based on interaction items in the time sequence of historical user interaction items in the multiple domains within the corresponding time period, wherein the target domain is any domain among the multiple domains, and any hyperedge in the local hypergraph and the global hypergraph represents an interaction item time sequence subsequence constituted by interaction items of any user within the corresponding time period; and the local hypergraph is used to generate a hypergraph based on interaction items in the time sequence of historical user interaction items in the target domain within the corresponding time period. The local hypergraph and the global hypergraph are input into a cross-domain sequence recommendation model to predict the user's to-be-interacted items; the recommendation information corresponding to the to-be-interacted items is sent to the client device corresponding to the user, and the cross-domain sequence recommendation model is used to extract the collaboration information between the interactive item time subsequences of different users and within the interactive item time subsequences based on the local hypergraph and the global hypergraph through a hypergraph convolutional network, obtain the item features of the interactive items containing the collaboration information, determine the user-item interaction features of the user based on the item features corresponding to the interactive items in the interactive item time subsequence of any user in multiple time periods, and predict the to-be-interacted items based on the user-item interaction features; The client device is used to receive the recommendation information and display the items to be interacted with through a user interaction interface.

12. A computing device, It is characterized in that include: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer program / instructions are executed by the processor, the steps of the cross-domain sequence recommendation method based on the hypergraph convolutional neural network described in any one of claims 1 to 8 and / or the cross-domain sequence recommendation model training method described in claim 9 are implemented.

13. A computer-readable storage medium, It is characterized in that It stores a computer program / instruction, which, when executed by a processor, implements the steps of the cross-domain sequence recommendation method based on a hypergraph convolutional neural network as described in any one of claims 1 to 8, and / or the cross-domain sequence recommendation model training method as described in claim 9.

14. A computer program product, It is characterized in that It includes a computer program / instruction, which, when executed by a processor, implements the steps of the cross-domain sequence recommendation method based on a hypergraph convolutional neural network as described in any one of claims 1 to 8, and / or the cross-domain sequence recommendation model training method as described in claim 9.

Citation Information

Patent Citations

  • Service modeling and recommendation method based on category perception graph neural network

    CN114817663A

  • Recommendation model training method and device based on sequence diagram and hypergraph

    CN116541716A

  • Personalized recommendation method based on time sequence and cross-domain collaboration

    CN118069938A

Cited By

  • Multi-scene session type recommendation method based on large language model agent framework

    CN119988743A

  • Medicine combination recommendation method and system based on hierarchical hypergraph convolutional network

    CN120108778A