CN116894122B - Cross-view contrast learning group recommendation method based on hypergraph convolutional network - Google Patents

Abstract

The invention provides a cross-view contrast learning group recommendation method based on a hypergraph convolutional network. The method designs a multi-view framework which is a member level preference network view of hypergraph representation, a group level preference network view of overlap graph representation and an item level preference network view of bipartite graph representation respectively. For each data view, a particular graph structure is applied to encode behavior data to generate a group representation of the corresponding view. The hypergraph learning architecture provided by the invention learns aggregation of member levels and captures high-order collaborative information. Compared with the existing aggregation method, the aggregation method is carried out by means of hypergraph convolution, and different group preference is carried out to transfer information along the hyperedges. The method mines the preference of the group to the articles in a mode of constructing multiple views, so that scoring prediction work is accurately carried out.

Description

A cross-view comparison learning group recommendation method based on hypergraph convolutional network

Technical field

The present invention relates to the technical field of cross-view comparison learning group recommendation using a hypergraph convolutional network for preference prediction, and specifically relates to a cross-view comparison learning group recommendation method based on a hypergraph convolutional network.

Background technique

With the development of the Internet and the popularity of online community activities, people with similar backgrounds (such as hobbies, occupations, ages) form a fixed or temporary group to participate in different activities according to different needs. For example, users are divided into many interest groups according to their different interests, such as game groups, painting groups, etc., in order to obtain various activity resources. People also often get together for various group activities, such as temporary travel groups, team dinners or watching movies. Conventionally, these people may be familiar with each other, such as living together in a family; they may also be strangers to each other, meeting by chance in an activity, such as several travelers joining a tour group together. In these scenarios, we need to recommend one or several suitable projects for the group to meet the needs of the group. However, there are many users in each group, and there are individual differences in preferences among different users. Therefore, the ultimate purpose of group recommendation is to aggregate the different preferences of group members and recommend suitable and satisfactory items to the group. Group recommendation can not only save time in group decision-making, but also reduce unnecessary conflicts among group members.

Most existing methods use heuristic methods or methods based on attention mechanisms to aggregate the personal preferences of group members to infer the group's preferences. However, these methods only model the user preferences of a single group, ignoring the complex high-level interactions within and outside the group. Second, a group's final decision does not necessarily come from the preferences of group members. But existing methods are insufficient to model such cross-group preferences. In addition, due to the sparse group-item interactions, group recommendation suffers from the problem of data sparseness. If the above problems are not solved, the accuracy of the recommended results will be reduced.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies in the existing technology and thereby propose a cross-view comparison learning group recommendation method based on a hypergraph convolutional network. The method achieves the recommendation of multiple items with the highest ratings to the group, and in real life, suitable and satisfactory items need to be recommended to the group.

The present invention is realized through the following technical solutions. The present invention proposes a cross-view comparison learning group recommendation method based on a hypergraph convolution network. The method includes the following steps:

Step 1. Obtain the group interaction data set from CAMRa2011 and Mafengwo platform. The data set contains the interaction history of users to items, groups to items, and user-group composition relationships;

Step 2. The user set in the training set is U, U={u ₁ ,u ₂ ,…,u _h ,…,u _M }, h∈{1,…,M}, where u _h is the h-th user, M is the number of users; the product set is I, I={i ₁ , i ₂ ,..., j _j ,..., i _n }, j∈{1,..., N}, where i _j is the j-th commodity, N is the number of commodities; the group set is G, G={g ₁ , g ₂ ,..., g _t ,..., g _k }, t∈{1,... , k}, where g _t is the t-th group, k is the number of groups; among them, the t-th group g _t ∈G consists of a group of group members, and G(t)={u ₁ , u ₂ ,...,u _h ,..., _up } means, where u _h ∈U, p is the number of group members included in group g _t , G(t) is the set of members of group g _t ;

Step 3. Construct a hypergraph with rich edge information, and expand the graph structure by connecting hyperedges of two or more nodes; among them, hyperedges can connect any number of nodes; the hypergraph is expressed as G ^m = (V ^m , ε ^m ), where V ^m =U∪I is a node set containing N unique vertices, each node represents a group member or an item of group interaction, ε ^m is an edge set containing M hyperedges, each hyperedge represents A group is composed of members in the group and items that interact with the group; formally, use ε ^t = {u ₁ , u ₂ ,…u _h …, _up , i ₁ , i ₂ ,…, i _j ,..., i _q } to represent the group g _t ; where, u _h ∈U, i _j ∈I, and ε ^t ∈ε ^m ; the connectivity of the hypergraph is represented by the correlation matrix to represent; for each vertex and hyperedge, use diagonal matrices D and B to represent the degrees of the vertex and hyperedge respectively, where/> Each hyperedge e∈ε contains two or more vertices and is assigned a positive weight W _ee , and all weights form a diagonal matrix W∈R ^M×M ;

Step 4. In the graph convolutional network on the overlapping graph of the hypergraph, connect similar groups to capture and propagate group-level preferences, and construct an overlapping graph; where, represented by G ^g = (V ^g , ε ^g ) The overlapping graph of the hypergraph; V ^g = {e: e∈ε}, ε ^g = {(e _p , e _q ): e _p , e _q ∈ε, |e _p ∩e _q |≥1}, and is Each edge in the overlapping graph is configured with a weight W _{p, q} , where W _{p, q} = |e _p ∩e _q |/|e _p ∪e _q |;

Step 5. Use the group-item bipartite graph to construct the graph G ^I =(V ^I , ε ^I ); where V ^I =G∪I represents the node set, ε ^I ={(g _t , i _j )g _t ∈ G, i _j ∈I, R(t, j)=1}; adjacency matrix

Step 6: Obtain group preferences by aggregating the preferences of members in the group from the member level using a hypergraph Capture and propagate group preferences from similar groups by leveraging overlay graphs/> Capture group preferences from the group’s interaction history by leveraging the group-item bipartite graph/> Three different gatings are used to automatically distinguish the contributions of different views, and the final group representation g _t is calculated:/> where α, β and γ respectively represent the learned weights, respectively represented by and/> Obtain; where W ^M , W ^I and W ^G ∈R ^d are three different trainable weights, and σ is the activation function;

Step 7. Calculate the prediction score of group g _t for item i _j Arrange the scores in descending order to get a list of items recommended for the group; randomly select (g _t ,i _j ) from R and sample negative samples for each group g _t , and use pairwise loss to calculate the group prediction loss, specifically As follows:/> Among them, O _G = {(t,j,j')|(t,j)∈O _G+ ,(t,j')∈O _G- } represents the group-item training data set, and O _G+ is the observed The set of interactions, O _G- is the set of unobserved interactions;

Step 8. Model the cross-view collaborative association and establish a cross-view contrast loss; use the obtained three group preference representations to obtain the contrast loss L _con ; combine the group recommendation loss and the contrast loss for joint training to minimize the following goals Function to learn model parameters: L=L _group +λL _con ; λ is a hyperparameter that controls contrast loss.

Further, in step 3, a member-level preference network is constructed and a hypergraph convolution operation is performed to encode the high-order relationship between users and items; the user-item aggregation process is M ^(l+1) =D ^-1 HWB ^-1 H ^T M ^(l) Θ ^(l) , where D, B and W represent the node degree matrix, edge degree matrix and weight matrix respectively; initialize the weight matrix W with the identity matrix so that all hyperedges have equal weight, Θ is a learnable parameter matrix between two convolutional layers; hypergraph convolution can be seen as two stages of information aggregation, "node-hyperedge-node"; that is, and

Further, in step 3, the attention mechanism is applied to learn the weights of members in the group; Among them, the weight α(h,j) represents the influence score of group member u _h when the group decides on project i _j . By calculating o(h,j)=h ^T RELU(Wu[u _h ;u' _h ] +Wj[i _j ; i' _j ]+b) and then softmax normalization.

Further, in step 4,

Input the group embedding G∈R ^k×d into the graph convolution network, denoted as G ⁽⁰⁾ = G, and perform the group-level graph convolution process Among them,/> I is the identity matrix, A _p,q = W _p,q ;/> is the diagonal matrix of the adjacency matrix, />

The group embeddings obtained at each layer are averaged to obtain the final group-level group embedding: Therefore, the group at the group level for each group g _t is expressed as/>

Further, in step 5,

The group embedding G∈R ^k×d and the item embedding I∈R ^n×d are sent to the graph convolution network, denoted as E ⁽⁰⁾ = E, where E is the concatenation of the two embeddings E = [G; I ]; perform item-level graph convolution:

The final group representation is obtained by averaging the representations learned at different layers, which is expressed as Obtain the project-level representation of each group g _t />

Further, in step 8,

Contrastive learning is applied on multiple views. For a node in one view, the same node embedding learned in another view is regarded as a positive sample pair; in both views, the node embeddings other than it are regarded as negative sample pairs; that is: Positive samples have one source, and negative samples have two sources, namely intra-view nodes and inter-view nodes.

Further, in step 8,

For the defined positive and negative samples, the contrast loss between the member-level preference view and the group-level preference view is Among them, the θ(·) function learns the score between the two input vectors and assigns a higher score to the positive sample compared to the negative sample pair. Specifically, use To calculate, h(·) is a nonlinear projection used to improve representation quality, mainly implemented by a two-layer perceptron; the contrast loss between the member-level preference view and the item-level preference view is The contrast loss between the group-level preference view and the item-level preference view is/>

Further, in step 8,

Since any two views are symmetrical, the calculation methods of L _GM , L _IM , and L _IG are the same as those of L _MG , L _MI , and L _GI . The final comparison loss between the member-level preference network view and the group-level preference network view is for In addition, the loss between any two views is calculated in the same way to obtain L _con2 and L _con3 ; then, the contrast losses of the three views are averaged to obtain the final contrast loss L _con :/>

The present invention proposes an electronic device, including a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it implements the cross-view comparison learning group based on a hypergraph convolution network. Recommended method steps.

The present invention proposes a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps of the cross-view comparison learning group recommendation method based on a hypergraph convolution network are implemented.

The invention has the following beneficial effects:

The present invention proposes a cross-view comparative learning hypergraph convolutional network model for group recommendation, abbreviated as "C ² -HGR". Mining group preferences for items by building multi-views to accurately predict ratings.

The present invention designs a multi-view learning framework at different granularity levels, including a member-level preference network represented by a hypergraph, a group-level preference network represented by an overlapping graph, and an item-level preference network represented by a bipartite graph. Through the effective integration of the three, user-project, group-project collaborative information and group similarity are extracted, thereby enhancing group preferences.

The present invention designs a new hypergraph neural convolutional network to obtain member-level aggregation, and uses the overlapping graph of hypergraph transformation to obtain group-level preferences. Compared with existing polymerization methods, the method of the present invention exhibits superior performance in terms of performance. Furthermore, in order to integrate group preference representations obtained from multiple views, the present invention designs an effective gating component to weigh the contribution of each view to the entire model.

The present invention proposes a method based on self-supervised multi-view contrastive learning to enhance group representation and solve the problem of data sparsity. This method is seamlessly coupled with the design of graph convolutional network layering. By unifying recommendation tasks and comparative learning tasks, recommendation performance can be significantly improved. And the present invention is suitable for group recommendation.

Description of the drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only These are embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on the provided drawings without exerting creative efforts.

Figure 1 is an overall schematic diagram of a cross-view comparative learning group recommendation method based on a hypergraph convolutional network according to the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

Combined with Figure 1, the present invention proposes a cross-view comparative learning group recommendation method based on hypergraph convolutional network. The method includes the following steps:

Step 1. Obtain the group interaction data set from CAMRa2011 and Mafengwo platform. The data set contains the interaction history of users to items, groups to items, and user-group composition relationships;

Step 2. The user set in the training set is U, U={u ₁ ,u ₂ ,…,u _h ,…,u _M }, h∈{1,…,M}, where u _h is the h-th user, M is the number of users; the product set is I, I={i ₁ ,i ₂ ,…,i _j ,…,i _n }, j∈{1,…,N}, where i _j is the j-th product, N is the number of commodities; the group set is G, G={g ₁ , g ₂ ,..., g _t ,..., g _k }, t∈{1,..., k}, where g _t is the t-th group, k is the number of groups; among them, the t-th group g _t ∈G consists of a group of group members, and G(t)={u ₁ , u ₂ ,...,u _h ,..., u _p } represents, where u _h ∈U, p is the number of group members contained in group g _t , and G(t) is the set of members of group g _t ;

Step 3. In order to capture complex and high-order group preferences, a hypergraph with rich edge information is planned to be constructed, and the graph structure is expanded by hyperedges connecting more than two nodes; among them, hyperedges can connect any number of nodes; super The graph is represented as G ^m =(V ^m , ε ^m ), where V ^m =U∪I is a node set containing N unique vertices, each node represents a group member or an item of group interaction, and ε ^m is a node set containing M An edge set of hyperedges, each hyperedge represents a group, which is composed of members in the group and items that the group interacts with; formally, use ε ^t = {u ₁ , u ₂ ,…u _h … , u _p , i ₁ , i ₂ ,..., i _j ,..., i _q } to represent the group g _t ; among them, u _h ∈U, i _j ∈I, and ε ^t ∈ε ^m ; the connectivity of the hypergraph correlation matrix to represent; for each vertex and hyperedge, use diagonal matrices D and B to represent the degrees of the vertex and hyperedge respectively, where Each hyperedge e∈ε contains two or more vertices and is assigned a positive weight W _ee , and all weights form a diagonal matrix W∈R ^M×M ;

Step 4. In the graph convolutional network on the overlapping graph of the hypergraph, the overlapping graph is constructed by connecting similar groups to capture and propagate group-level preferences; where G ^g = (V ^g , εg) represents the super graph. The overlapping graph of the graph; V ^g = {e: e∈ε}, ε ^g = {(e _p , e _q ): e _p , e _q ∈ε, |e _p ∩e _q |≥1}, and is overlapping Each edge in the graph is configured with a weight W _{p, q} , where W _{p, q} = |e _p ∩e _q |/|e _p ∪e _q |;

Step 5. Use the group-item bipartite graph to construct the graph G ^I =(V ^I , ε ^I ); where V ^I =G∪I represents the node set, ε ^I ={(g _t , i _j )|g _t ∈G, i _j ∈I, R(t, j)=1}; adjacency matrix

Step 6: Obtain group preferences by aggregating the preferences of members in the group from the member level using a hypergraph Capture and propagate group preferences from similar groups by leveraging overlay graphs/> Capture group preferences from the group’s interaction history by leveraging the group-item bipartite graph/> Three different gatings are used to automatically distinguish the contributions of different views, and the final group representation g _t is calculated:/> where α, β and γ respectively represent the learned weights, respectively represented by and/> Obtain; where W ^M , W ^I and W ^G ∈R ^d are three different trainable weights, and σ is the activation function;

Step 7. Calculate the prediction score of group g _t for item i _j Arrange the scores in descending order to get a list of items recommended for the group; randomly select (g _t ,i _j ) from R and sample negative samples for each group g _t , and use pairwise loss to calculate the group prediction loss, specifically As follows:/> Among them, O _G = {(t,j,j')|(t,j)∈O _G+ ,(t,j')∈O _G- } represents the group-item training data set, and O _G+ is the observed The set of interactions, O _G- is the set of unobserved interactions;

Step 8. Model the cross-view collaborative association and establish a cross-view contrast loss; use the three group preference representations obtained in step 4 to obtain the contrast loss L _con ; combine the group recommendation loss and the contrast loss for joint training, and the minimum The following objective function is used to learn model parameters: L=L _group +λL _con ; λ is a hyperparameter that controls contrast loss.

In step 3, a member-level preference network is constructed and a hypergraph convolution operation is performed to encode the high-order relationship between users and items; the user-item aggregation process is M ^(l+1) =D ^-1 HWB ^-1 H ^T M ^(l) Θ ^(l) , where D, B and W represent the node degree matrix, edge degree matrix and weight matrix respectively; initialize the weight matrix W with the identity matrix so that all hyperedges have equal weight, Θ is two A parameter matrix that can be learned between convolutional layers; specifically, hypergraph convolution can be seen as two stages of information aggregation, "node-hyperedge-node"; that is, and

In step 3, apply the attention mechanism to learn the weight of members in the group; specifically, Among them, the weight α(h,j) represents the influence score of group member u _h when the group decides on project i _j . By calculating o(h,j)=h ^T RELU(Wu[u _h ;u' _h ] +Wj[i _j ; i' _j ]+b) and then softmax normalization.

In step 4,

Input the group embedding G∈R ^k×d into the graph convolution network, denoted as G ⁽⁰⁾ = G, and perform the group-level graph convolution process Among them,/> I is the identity matrix, A _p,q = W _p,q ;/> is the diagonal matrix of the adjacency matrix, />

The group embeddings obtained at each layer are averaged to obtain the final group-level group embedding: Therefore, the group at the group level for each group g _t is expressed as/>

In step 5,

In order to capture the collaboration signal between groups and projects, the group embedding G∈R ^k×d and the project embedding I∈R ^n×d are sent to the graph convolution network, denoted as E ⁽⁰⁾ = E, where E is the concatenation of two embeddings E = [G; I]; perform item-level graph convolution:

The final group representation is obtained by averaging the representations learned at different layers, which is expressed as Obtain the project-level representation of each group g _t />

In step 8,

In order to solve the problem of sparse user-item and group-item interactions and refine user and group representations, contrastive learning is applied on multiple views. For a node in one view, the same node embedding learned in another view is regarded as a positive sample. Pair; in the two views, node embeddings other than it are regarded as negative sample pairs; that is: positive samples have one source, and negative samples have two sources, namely intra-view nodes and inter-view nodes.

In step 8,

For the defined positive and negative samples, the contrast loss between the member-level preference view and the group-level preference view is Among them, the θ(·) function learns the score between the two input vectors and assigns a higher score to the positive sample compared to the negative sample pair. Specifically, use To calculate, h(·) is a nonlinear projection used to improve representation quality, mainly implemented by a two-layer perceptron; the contrast loss between the member-level preference view and the item-level preference view is The contrast loss between the group-level preference view and the item-level preference view is/>

In step 8,

Since any two views are symmetrical, the calculation methods of L _GM , L _IM , and L _IG are the same as those of L _MG , L _MI , and L _GI . The final comparison loss between the member-level preference network view and the group-level preference network view is for In addition, the loss between any two views is calculated in the same way to obtain L _con2 and L _con3 ; then, the contrast losses of the three views are averaged to obtain the final contrast loss L _con :/>

The present invention proposes a cross-view comparison learning group recommendation method based on a hypergraph convolutional network. The method designs a multi-view framework, which are member-level preference network views represented by hypergraphs and group-level preferences represented by overlapping graphs. Network views and bipartite graph representations of item-level preference network views. For each data view, a specific graph structure is applied to encode the behavioral data and generate a group representation of the corresponding view. Among them, the hypergraph learning architecture proposed by the present invention learns member-level aggregation and captures high-order collaborative information. Compared with existing aggregation methods, this aggregation method relies on hypergraph convolution, and different group preferences transfer information along the hyperedge. For general preferences of groups, the proposed item-level preference network and group-level preference network are proposed. The two learn group representation through multi-layer convolution operations based on group-item interaction information and group similarity (i.e., the overlapping relationship between groups) respectively. Utilizing multi-view convolutional networks, a gating component is further proposed to adaptively adjust the contribution of each view. Secondly, in order to alleviate the problem of data sparseness, a contrastive learning method is proposed to be applied on multiple views. Optimize model parameters through unified recommendation tasks and comparative learning tasks to provide good decision-making results for the group.

The present invention proposes an electronic device, including a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it implements the cross-view comparison learning group based on a hypergraph convolution network. Recommended method steps.

The present invention proposes a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps of the cross-view comparison learning group recommendation method based on a hypergraph convolutional network are implemented.

The memory in the embodiment of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory can be read only memory (ROM), programmable ROM (PROM), erasable programmable read-only memory (erasablePROM, EPROM), electrically erasable memory Programmable read-only memory (electrically EPROM, EEPROM) or flash memory. Volatile memory may be random access memory (RAM), which is used as an external cache. By way of illustration, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM) ), double data rate synchronous dynamic random access memory (double data rate SDRAM, DDRSDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (synchlink DRAM, SLDRAM) and Direct memory bus random access memory (direct rambusRAM, DRRAM). It should be noted that the memory of the method described herein is intended to include, but is not limited to, these and any other suitable types of memory.

In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on the computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in or transmitted from one computer-readable storage medium to another, e.g., the computer instructions may be transferred from a website, computer, server, or data center Transmission to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains one or more available media integrated therein. The usable media may be magnetic media (eg, floppy disks, hard disks, tapes), optical media (eg, high-density digital video discs (DVD)), or semiconductor media (eg, solid state discs, SSD)) etc.

During the implementation process, each step of the above method can be completed by instructions in the form of hardware integrated logic circuits or software in the processor. The steps of the methods disclosed in conjunction with the embodiments of the present application can be directly implemented by a hardware processor, or executed by a combination of hardware and software modules in the processor. The software module can be located in random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, registers and other mature storage media in this field. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

It should be noted that the processor in the embodiment of the present application may be an integrated circuit chip with signal processing capabilities. During the implementation process, each step of the above method embodiment can be completed through an integrated logic circuit of hardware in the processor or instructions in the form of software. The above-mentioned processor may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. . Each method, step and logical block diagram disclosed in the embodiment of this application can be implemented or executed. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present application can be directly implemented by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software module can be located in random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, registers and other mature storage media in this field. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

The above is a detailed introduction to the cross-view comparison learning group recommendation method based on hypergraph convolutional network proposed by the present invention. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above embodiments The description is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, there will be changes in the specific implementation and application scope based on the idea of the present invention. In summary, , the contents of this description should not be construed as limitations of the present invention.

Claims (10)

1. A cross-view comparative learning group recommendation method based on hypergraph convolutional network, characterized in that: the method includes the following steps: Step 1. Obtain the group interaction data set from CAMRa2011 and Mafengwo platform. The data set contains the interaction history of users to items, groups to items, and user-group composition relationships; Step 2. The user set in the training set is U, U={u ₁ , u ₂ ,..., u _h ,..., u _M }, h∈{1,..., M}, where u _h is the h-th user, M is the number of users; the product set is I, I={i ₁ , i ₂ ,..., i _j ,..., i _n }, j∈{1,..., N}, where i _j is the j-th commodity, N is the quantity of the commodity; the group set is G, G={g ₁ , g ₂ ,..., g _t ,..., g _k }, t∈ {1,...,k}, where g _t is the t-th group and k is the number of groups; among them, the t-th group g _t ∈G consists of a group of group members, and G(t)= {u ₁ , u ₂ , ..., u _h , ..., u _p } means, where u _h ∈U, p is the group g _t contains the number of group members, and G(t) is the group g A collection of _t members; Step 3. Construct a hypergraph with rich edge information, and expand the graph structure by connecting hyperedges of two or more nodes; among them, hyperedges can connect any number of nodes; the hypergraph is expressed as G ^m = (V ^m , ε ^m ), where V ^m =U∪I is a node set containing N unique vertices, each node represents a group member or an item of group interaction, ε ^m is an edge set containing M hyperedges, each hyperedge represents A group is composed of members in the group and items that interact with the group; formally, use ε ^t = {u ₁ , u ₂ ,…u _h …, _up , i ₁ , i ₂ ,…, i _j ,..., i _q } to represent the group g _t ; where, u _h ∈U, o _j ∈I, and ε ^t ∈ε ^m ; the connectivity of the hypergraph is represented by the correlation matrix to represent; for each vertex and hyperedge, use diagonal matrices D and B to represent the degrees of the vertex and hyperedge respectively, where/> Each hyperedge e∈ε contains two or more vertices and is assigned a positive weight W _ee , and all weights form a diagonal matrix W∈R ^M×M ; Step 4. In the graph convolutional network on the overlapping graph of the hypergraph, connect similar groups to capture and propagate group-level preferences, and construct an overlapping graph; where, represented by G ^g = (V ^g , ε ^g ) The overlapping graph of the hypergraph; V ^g = {e: e∈ε}, ε ^g = {(e _p , e _q ): e _p , e _q ∈ε, |e _p ∩e _q |≥1}, and is Each edge in the overlapping graph is configured with a weight W _{p, q} , where W _{p, q} = |e _p ∩e _q |/e _p ∪e _q |; Step 5. Use the group-item bipartite graph to construct the graph G ^I =(V ^I , ε ^I ); where V ^I =G∪I represents the node set, ε ^I ={(g _t , i _j )|g _t ∈G, i _j ∈I, R(t, j)=1}; adjacency matrix Step 6: Obtain group preferences by aggregating the preferences of members in the group from the member level using a hypergraph Capture and propagate group preferences from similar groups by leveraging overlay graphs/> Capturing group preferences from the group’s interaction history by leveraging group-item bipartite graphs/> Three different gatings are used to automatically distinguish the contributions of different views, and the final group representation g _t is calculated:/> where a, β and γ respectively represent the learned weights, respectively represented by and/> Obtain; where W ^M , W ^I and W ^G ∈R ^d are three different trainable weights, and σ is the activation function; Step 7. Calculate the prediction score of group g _t for item i _j Arrange the scores in descending order to get a list of items recommended for the group; randomly select (g _t , i _j ) from R and sample negative samples for each group g _t , and use pairwise loss to calculate the group prediction loss, specifically As follows:/> in, Represents the group-item training data set, /> is the set of observed interactions, /> is a collection of unobserved interactions; Step 8. Model the cross-view collaborative association and establish a cross-view contrast loss; use the obtained three group preference representations to obtain the contrast loss L _con ; combine the group recommendation loss and the contrast loss for joint training to minimize the following goals Function to learn model parameters: L=L _group +λL _con ; λ is a hyperparameter that controls contrast loss. 2. The method according to claim 1, characterized in that: in step 3, constructing a member-level preference network and performing a hypergraph convolution operation to encode high-order relationships between users and items; user-item aggregation process is M ^(l+1) =D ^-1 HWB ^{- 1} H ^T M ^(l) Θ ⁽¹⁾ , where D, B and W represent the node degree matrix, edge degree matrix and weight matrix respectively; initialize the weights with the unit matrix The matrix W makes all hyperedges have equal weight, and Θ is the parameter matrix that can be learned between the two convolution layers; hypergraph convolution can be seen as two stages of information aggregation, "node-hyperedge-node"; Right now and/> 3. The method according to claim 2, characterized in that: in step 3, an attention mechanism is applied to learn the weight of members in the group; Among them, the weight α (h, j) represents the influence score of group member u _h when the group makes decision on project i _j . By calculating o (h, j) = h ^T RELU (Wu [u _h ; u' _h ] +Wj[i _j ; i' _j ]+b) and then softmax normalization. 4. The method according to claim 3, characterized in that: in step 4, Input the group embedding G∈R ^k×d into the graph convolution network, denoted as G ⁽⁰⁾ = G, and perform the group-level graph convolution process Among them,/> I is the identity matrix, A _{p, q} = W _{p, q} ;/> is the diagonal matrix of the adjacency matrix, /> The group embeddings obtained at each layer are averaged to obtain the final group-level group embedding: Therefore, the group at the group level for each group g _t is expressed as/> 5. The method according to claim 4, characterized in that: in step 5, The group embedding G∈R ^k×d and the item embedding I∈R ^n×d are sent to the graph convolution network, denoted as E ⁽⁰⁾ = E, where E is the concatenation of the two embeddings E = [G; I ]; perform item-level graph convolution: The final group representation is obtained by averaging the representations learned at different layers, which is expressed as Obtain the project-level representation of each group g _t /> 6. The method according to claim 1, characterized in that: in step 8, Contrastive learning is applied on multiple views. For a node in one view, the same node embedding learned in another view is regarded as a positive sample pair; in both views, the node embeddings other than it are regarded as negative sample pairs; that is: Positive samples have one source, and negative samples have two sources, namely intra-view nodes and inter-view nodes. 7. The method according to claim 6, characterized in that: in step 8, For the defined positive and negative samples, the contrast loss between the member-level preference view and the group-level preference view is Among them, the θ(·) function learns the score between the two input vectors and assigns a higher score to the positive sample compared to the negative sample pair. Specifically use/> To calculate, h(·) is a nonlinear projection used to improve representation quality, mainly implemented by a two-layer perceptron; the contrast loss between the member-level preference view and the item-level preference view is/> The contrast loss between the group-level preference view and the item-level preference view is/> 8. The method according to claim 7, characterized in that: in step 8, Since any two views are symmetrical, the calculation methods of L _GM , L _IM , and L _IG are the same as those of L _MG , L _MI , and L _GI . The final comparison loss between the member-level preference network view and the group-level preference network view is for In addition, the loss between any two views is calculated in this way to obtain L _con2 and L _con3 ; then, the contrast losses of the three views are averaged to obtain the final contrast loss Lcon:/> 9. An electronic device, including a memory and a processor, the memory stores a computer program, characterized in that when the processor executes the computer program, the steps of the method of any one of claims 1-8 are implemented. 10. A computer-readable storage medium used to store computer instructions, characterized in that when the computer instructions are executed by a processor, the steps of the method described in any one of claims 1-8 are implemented.