A multi-layer information interaction recommendation method based on a gated graph neural network

Abstract

The application discloses a kind of multi-layer information interaction recommendation method based on gated graph neural network, and its steps include:1, from online social platform, the three kinds of binary set data set including user, group, project are collected;2, the following three network graphs are constructed: user social relationship graph, project correlation graph, user-project interaction graph;3, the gated graph neural network of multi-layer information interaction is constructed, and model parameters are optimized;4, the first H item with the highest prediction score is selected and recommended to user.The application can utilize the user social relationship in group and project correlation to construct weighted graph neural network, to alleviate the problem of data scarcity and can improve recommendation accuracy and personalized degree, can be used in the item recommendation in various social websites, with strong generalization ability, wide application range, with higher application value.

Description

Technical Field

[0001] This invention belongs to the field of recommendation system technology, specifically relating to a multi-layer information interaction recommendation method based on gated graph neural networks. Background Technology

[0002] With the increasing popularity of social networking sites and the booming development of information technology, people face a serious problem of information overload while enjoying these conveniences. Recommendation systems can effectively alleviate this problem, and many recommendation methods have been proposed to help users find items they may be interested in. Existing recommendation methods are mainly divided into traditional methods and deep learning methods. Deep learning methods, due to their powerful representation learning capabilities, can overcome many limitations of traditional methods and have attracted much attention from researchers. Among them, graph neural networks have achieved remarkable success in analyzing graph data, and user-item interactions can be naturally constructed as graph data. Therefore, the application of graph neural networks presents an unprecedented challenge to solving the recommendation problem.

[0003] However, existing graph neural network-based recommendation methods still have shortcomings in solving recommendation problems in social networks:

[0004] (1) Most existing recommendation methods only consider the interaction information between users and items, resulting in data sparsity and cold start problems. Currently, there are a large number of interest groups on social networking sites, which can be used to explore the social influence relationships between users and the similarity relationships between items. Using this information can further improve the accuracy and personalization of recommendation systems.

[0005] (2) Some recommendation methods based on graph neural networks take into account friendship relationships or similarity between items when making recommendations. However, these methods do not consider the influence between users or the degree of similarity between items when constructing the graph neural network. Ignoring these differences will prevent the effective capture of user preferences and item features, ultimately affecting the recommendation results. Summary of the Invention

[0006] To address the aforementioned problems in existing technologies, this invention provides a multi-layer information interaction recommendation method based on gated graph neural networks. This method aims to extract different user preferences and item features from various aspects and fuse them through a gating mechanism, thereby improving the recommendation effect.

[0007] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0008] The present invention provides a multi-layer information interaction recommendation method based on a gated graph neural network, characterized by the following steps:

[0009] Step 1: Collect a user set U = {u1, u2, ..., u...} from online social platforms.m ,…,u M}, Project set V = {v1, v2, ..., v n ,…,v N} and group collection Among them, u m Let v represent the m-th user. n This represents the nth item. Let G represent the k-th group, M represent the number of users, N represent the number of items, and K represent the number of groups. In the group set G, each group is a set consisting of a certain number of users and items.

[0010] Step 2, construct three network diagrams, including: a user-project interaction diagram, a user social relationship diagram, and a project-related relationship diagram:

[0011] Step 2.1: Construct a user-project interaction graph G based on the direct interaction relationship between users and projects. t (U∪V,E), where E represents the set of edges representing user-item interaction relationships. If the m-th user u m With the nth item v n If an interaction occurs, then the m-th user u m and the nth item v n There is an edge between them;

[0012] Step 2.2: Construct a user social relationship graph G based on the social relationships between users in the user set U. u (U,S,X), where... Represents the m-th user u m and the c-th user u c Do they belong to the kth group? Let X represent the social relationships, and let X be the weighted neighbor matrix. The element X in the m-th row and c-th column of X is obtained by equation (1). mc :

[0013]

[0014] In equation (1), and Representing the m-th user u m The set of groups joined and the c-th user u c The group collection to join

[0015] Step 2.3: Based on the relationships between the projects in the project set V, construct the project relationship graph G. v (V,C,Y), where... v represents the s-th item s and the tth project v tWhether or not they are shared in the kth group The correlation is given by Y, which represents the weighted neighbor matrix, and the element Y in the s-th row and t-th column of Y is obtained by equation (2). st :

[0016]

[0017] In equation (2), and These represent the s-th project v that has been shared. s The group set and the t-th item v t A collection of groups,

[0018] Step 3: Construct a multi-layer gated graph neural network for information interaction, including: an embedding layer, an interaction graph information propagation layer, a relationship graph information propagation layer, a gating layer, and a prediction layer.

[0019] Step 3.1, the embedding layer for the m-th user u m and the nth project n n Preprocessing is performed to obtain the m-th user u m Initial user embedding vector and the nth item v n Initial item embedding vector

[0020] Step 3.2, the interactive graph information propagation layer respectively... and G t (U∪V,E) and and G t Process (U∪V,E) to obtain the m-th user u. m In G t User interaction information aggregation embedding in layer l of (U∪V,E) and the nth item v n In G t The project interaction information aggregation embedding in the l-th layer of (U∪V,E)

[0021] Step 3.3, the relationship graph information propagation layer respectively... and G u (U,S,X) and and G v Process (V,C,Y) to obtain the m-th user u. m In G u User relationship information aggregation embedding in layer l of (U,S,X) and the nth item v n In the project relationship diagram G vThe project relationship information aggregation embedding in the l-th layer of (V,C,Y)

[0022] Step 3.4, the gating layer... and Perform fusion to obtain the m-th user u m The l-th layer user embedding vector and the nth item v n The l-th layer item embedding vector

[0023] Step 3.5: Obtain the m-th user u following the process from Steps 3.2 to 3.4. m Lth layer user embedding vector and the nth item v n The Lth layer of the project embedding vector layer Thus predicting the m-th user u m For the nth item v n Preference score

[0024] Step 3.6: With the goal of minimizing the BPR loss function Loss, the parameters of the gated graph neural network with multi-layer information interaction are optimized and updated through the backpropagation algorithm and the Adam algorithm until the number of iterations reaches the set upper limit, thereby obtaining the optimal multi-layer information interaction model after training.

[0025] Step 4: Input the new set of users and items to be recommended into the optimal multi-level information interaction model for processing, and output the preference score of each user for each item. Then, select the top H items with the highest predicted score for each user and which have not been interacted with to form each recommendation list and recommend them to the corresponding users.

[0026] The feature of the multi-layer information interaction recommendation method based on gated graph neural networks described in this invention is that step 3.2 includes the following steps:

[0027] Step 3.2.1, use equation (3) to calculate the m-th user u m Its neighboring project node v b The attention coefficient of the l-th layer

[0028]

[0029] In equation (3), It is the m-th user u m In G t Neighboring project nodes in (U∪V,E), N t (u m ) represents G tThe m-th user u in (U∪V,E) m The set of neighboring project nodes, and Representing the m-th user u m and v b In G t User embedding vector and item embedding vector in layer (l-1) of (U∪V,E) Indicates in G t In (U∪V,E), the neighboring item node v at layer l is... b For the m-th user u m The importance of f, where f represents a single-layer feedforward neural network. G represents u In (U,S,X), the shared weight matrix of the l-th layer, [·||·] represents the concatenation operation;

[0030] Step 3.2.2, use equation (4) to obtain G t In (U∪V,E), the neighboring item node v at layer l is... b For the m-th user u m Attention coefficient after regularization

[0031]

[0032] In equation (4), G represents t Another neighboring item node in layer l of (U∪V,E) For the m-th user u m Attention coefficient;

[0033] Step 3.2.3, use equation (5) to obtain the m-th user u m In the user-project interaction graph G t User interaction information aggregation embedding in layer l of (U∪V,E)

[0034]

[0035] In equation (5), σ(·) represents the Sigmoid function. Represents the neighboring project node v b In G t The item embedding vector of the (l-1)th layer in (U∪V,E);

[0036] Step 3.2.4: Following the process from steps 3.3.1 to 3.3.3, obtain the nth item v. n In the user-project interaction graph G t The project interaction information aggregation embedding in the l-th layer of (U∪V,E)

[0037] Step 3.3 includes the following steps:

[0038] Step 3.3.1, use equation (6) to calculate the m-th user u m Its neighboring user node u c Normalized edge weights a mc :

[0039]

[0040] In equation (6), Represents the m-th user u m In the user social relationship graph G u Degree in (U,S,X) Represents the m-th user u m In the user social relationship graph G u The set of neighboring users in (U,S,X);

[0041] Step 3.3.2, use equation (7) to obtain the m-th user u m In the user social relationship graph G u User relationship information aggregation embedding in layer l of (U,S,X)

[0042]

[0043] In equation (7), G represents u The shared weight matrix of the l-th layer in (U,S,X), where σ(·) represents the ReLU activation function, b l G represents u In (U,S,X), the bias parameters to be learned in the l-th layer are given, where L represents the optimal number of network layers. Represents the m-th user u m The (l-1)th layer user embedding vector;

[0044] Step 3.3.3: Obtain the nth item v following the process of steps 3.2.1-3.2.2. n In the project relationship diagram G v The project relationship information aggregation embedding in the l-th layer of (V,C,Y)

[0045] Step 3.4 involves using equations (8) and (9) to calculate... and

[0046]

[0047]

[0048] In equations (8) and (9), ω1 and ω2 represent two vectors to be learned, and σ(·) represents the Sigmoid function. It represents the Hadamardi (or Hadama) stack;

[0049] In step 3.6, the BPR loss function Loss is constructed using equation (10):

[0050]

[0051] In equation (10), Represents the m-th user u m The sample set consisting of items in the item set V, n + Indicates the relationship with the m-th user u m Projects that have generated interaction, n - This indicates that a random sample is taken from the project set V and the m-th user u is selected. m Projects that have not generated any interaction This represents the set of positive samples consisting of user-project interactions. Let represent the set of negative user-item interactions randomly sampled from user set U and item set V; σ(·) represents the Sigmoid function, Θ represents all parameters to be trained in the model, and λ is the regularization coefficient.

[0052] The present invention provides an electronic device, comprising a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the multi-layer information interaction recommendation method, and the processor is configured to execute the program stored in the memory.

[0053] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program is executed by a processor to perform the steps of the multi-layer information interaction recommendation method.

[0054] Compared with existing technologies, the beneficial effects of this invention are reflected in:

[0055] 1. This invention not only utilizes user-item interaction information, but also considers the social relationships between users and the relevance of items in a group to build a model for recommendation. It can overcome the problems of data sparsity and cold start, and also provides a reference for the utilization of group information in the research of item recommendation.

[0056] 2. This invention designs a weighted graph neural network to extract user and item representations from user social relationship graphs and item-related relationship graphs, which can more comprehensively capture user and item features and is beneficial to improving the accuracy and personalization of recommendation systems.

[0057] 3. This invention can be used to recommend various items such as movies, music, and papers on various social networking sites. It has strong generalization ability, wide applicability, and better practical application value. Attached Figure Description

[0058] Figure 1 This is a schematic flowchart of the method of the present invention;

[0059] Figure 2 This is a structural diagram of the multi-layer information interaction recommendation model based on gated graph neural network of the present invention. Detailed Implementation

[0060] In this embodiment, as Figure 1 As shown, a multi-layer information interaction recommendation method based on gated graph neural networks can construct a weighted graph neural network by utilizing the social relationships and item-related relationships of users in a group. This alleviates the problem of data scarcity and improves recommendation accuracy and personalization. It can be used for item recommendation on various social networking sites, has strong generalization ability, wide application range, and higher application value. Specifically, the method includes the following steps:

[0061] Step 1: Collect a user set U = {u1, u2, ..., u...} from online social platforms. m ,…,u M}, Project set V = {v1, v2, ..., v n ,…,v N} and group collection Among them, u m Let v represent the m-th user. n This represents the nth item. Let G represent the k-th group, M represent the number of users, N represent the number of items, and K represent the number of groups. In the group set G, each group is a set consisting of a certain number of users and items.

[0062] The dataset of this invention can obtain interaction data between users, projects, and groups from various existing online social networking sites (such as Douban, Goodreads, and ResearchGate). When preprocessing the binary data, users or projects with too few interactions should be filtered out, such as users who interact too little with projects or join too few groups.

[0063] Step 2, construct three network diagrams: a user-project interaction diagram, a user social relationship diagram, and a project-related relationship diagram.

[0064] Step 2.1: Construct a user-project interaction graph G based on the direct interaction relationship between users and projects. t (U∪V,E), where E represents the set of edges representing user-item interaction relationships. If the m-th user u mWith the nth item v n If an interaction occurs, then the m-th user u m and the nth item v n There is an edge between them.

[0065] Step 2.2: Construct a user social relationship graph G based on the social relationships between users in the user set U. u (U,S,X), where... Represents the m-th user u m and the c-th user u c Do they belong to the kth group? Let X represent the social relationships, and let X be the weighted neighbor matrix. The element X in the m-th row and c-th column of X is obtained by equation (1). mc :

[0066]

[0067] In equation (1), and Representing the m-th user u m The set of groups joined and the c-th user u c The group collection to join X mc This represents the social influence value between two users. The more common groups two users join, the more likely their interests and preferences are to be influenced by each other. Calculating this social influence value can better capture changes in users' interests and preferences.

[0068] Step 2.3: Construct a project relationship graph G based on the relationships between the projects in the project set V. v (V,C,Y), where... v represents the s-th item s and the t-th user v t Shared in the kth group The correlation is given by Y, which represents the weighted neighbor matrix, and the element Y in the s-th row and t-th column of Y is obtained by equation (2). st :

[0069]

[0070] In equation (2), and These represent the s-th project v that has been shared. s The group set and the t-th item v t A collection of groups, Y stThis indicates the correlation value between two projects. When two projects are always shared in certain groups, they are more likely to be related. Calculating this correlation value helps recommend projects that are more relevant to users' interests.

[0071] Step 3: Construct a multi-layer gated graph neural network for information interaction, including: an embedding layer, an interaction graph information propagation layer, a relationship graph information propagation layer, a gating layer, and a prediction layer.

[0072] Step 3.1: The embedding layer performs one-hot encoding on the user set U and the item set V, and then passes the encoded user vector and item vector through the multilayer perceptron to be trained to obtain the m-th user u. m Initial user embedding vector and the nth item v n Initial item embedding vector

[0073] Step 3.2, due to the user-project interaction graph G t It demonstrates direct interaction between users and projects, thus allowing inferences about user preferences from projects they have interacted with, and also inferences about project characteristics from users who have shown interest in projects. Therefore, the interaction graph information propagation layer respectively... and G t (U∪V,E) and and G t Process (U∪V,E) to obtain the m-th user u. m In G t User interaction information aggregation embedding in layer l of (U∪V,E) and the nth item v n In G t The project interaction information aggregation embedding in the l-th layer of (U∪V,E)

[0074] Step 3.2.1, in the user-project interaction diagram G t When aggregating neighbor node information, the contribution of each neighbor node is different. To capture this difference, equation (3) is used to calculate the neighbor item node v of the l-th layer. b For the m-th user node u m Attention coefficient

[0075]

[0076] In equation (3), It is the m-th user u m In G t Neighboring project nodes in (U∪V,E), Nt (u m ) represents G t The m-th user u in (U∪V,E) m The set of neighboring project nodes, and Representing the m-th user u m and v b In G t User embedding vector and item embedding vector in layer (l-1) of (U∪V,E) Indicates in G t In (U∪V,E), the neighboring item node v at layer l is... b For the m-th user u m The importance of f, where f represents a single-layer feedforward neural network. G represents u In (U,S,X), the shared weight matrix of the l-th layer, [·||·] represents the concatenation operation;

[0077] Step 3.2.2, use equation (4) to obtain G t In (U∪V,E), the neighboring item node v at layer l is... b For the m-th user u m Attention coefficient after regularization

[0078]

[0079] In equation (4), G represents t Another neighboring item node in layer l of (U∪V,E) For the m-th user u m Attention coefficient after regularization

[0080] Step 3.2.3, use equation (5) to obtain the m-th user u m In the user-project interaction graph G t User interaction information aggregation embedding in layer l of (U∪V,E)

[0081]

[0082] In equation (5), σ(·) represents the Sigmoid function. Represents the neighboring project node v b In G t The item embedding vector of the (l-1)th layer in (U∪V,E) This represents the m-th user u m Interest preferences regarding the project;

[0083] Step 3.2.4: Following the process from steps 3.3.1 to 3.3.3, obtain the nth item v. n In the user-project interaction graph G t The project interaction information aggregation embedding in the l-th layer of (U∪V,E) This represents the nth item v. n Features that will be of interest to users;

[0084] Step 3.3, User Social Relationship Graph G u This illustrates the varying strengths of social connections among users mediated by groups. Users' interests and preferences are influenced by those of users with whom they have social connections; therefore, it's necessary to consider these socially influenced user preferences. Similarly, the project-related relationship graph G... v This demonstrates varying degrees of correlation between items. A user is more likely to click on or browse items with higher relevance to their past interactions. Therefore, it's necessary to consider these correlations to better capture features that might interest users. The relationship graph information propagation layer respectively... and G u (U,S,X) and and G v Process (V,C,Y) to obtain the m-th user u. m In G u User relationship information aggregation embedding in layer l of (U,S,X) and the nth item v n In the project relationship diagram G v The project relationship information aggregation embedding in the l-th layer of (V,C,Y)

[0085] Step 3.3.1: Since the number of groups joined by different users varies greatly, the m-th user u is calculated using equation (6). m Its neighboring user node u c Normalized edge weights a mc :

[0086]

[0087] In equation (6), Represents the m-th user u m In the user social relationship graph G u The degree in (U,S,X) also represents the degree between the user set U and the m-th user u. m The number of users with social relationships Represents the m-th user node u m In the user social relationship graph G u The set of neighboring users in (U,S,X), amc Representing neighbor user u c For the m-th user u m Social influence weighting;

[0088] Step 3.3.2, use equation (7) to obtain the m-th user u m In the user social relationship graph G u User relationship information aggregation embedding in layer l of (U,S,X)

[0089]

[0090] In equation (7), G represents u The shared weight matrix of the l-th layer in (U,S,X), where σ(·) represents the ReLU activation function, b l G represents u In (U,S,X), the bias parameters to be learned in the l-th layer are given, where L represents the optimal number of network layers. Represents the m-th user u m The (l-1)th layer user embedding vector, Represents the m-th user u m User preference characteristics influenced by social factors.

[0091] Step 3.3.3: Obtain the nth item v following the process of steps 3.2.1-3.2.2. n In the project relationship diagram G v The project relationship information aggregation embedding in the l-th layer of (V,C,Y) The nth project v represents the project that takes into account the relationships between projects. n Features that will be of interest to users.

[0092] Step 3.4, the gating layer... and Perform fusion to obtain the m-th user u m The l-th layer user embedding vector and the nth item v n The l-th layer item embedding vector The calculation methods are shown in equations (8) and (9):

[0093]

[0094]

[0095] In equations (8) and (9), ω1 and ω2 represent two vectors to be learned, and σ(·) represents the Sigmoid function. It represents the Hadamardi (or Hadama) stack;

[0096] Step 3.5: Obtain the m-th user u following the process from Steps 3.2 to 3.4. m Lth layer user embedding vector and the nth item v n The Lth layer of the project embedding vector layer Thus predicting the m-th user u m For the nth item v n Preference score

[0097] Step 3.6, construct the BPR loss function Loss using equation (10):

[0098]

[0099] In equation (10), Represents the m-th user u m The sample set consisting of items in the item set V, n + Indicates the relationship with the m-th user u m Projects that have generated interaction, n - This indicates that a random sample is taken from the project set V and the m-th user u is selected. m Projects that have not generated any interaction This represents the set of positive samples consisting of user-project interactions. Let represent the set of negative user-item interactions randomly sampled from user set U and item set V; σ(·) represents the Sigmoid function, Θ represents all parameters to be trained in the model, and λ is the regularization coefficient.

[0100] With the goal of minimizing the BPR loss function Loss, the parameters of the gated graph neural network with multi-layer information interaction are optimized and updated through the backpropagation algorithm and the Adam algorithm until the number of iterations reaches the set upper limit, thereby obtaining the optimal multi-layer information interaction model after training.

[0101] Step 4: Input the new set of users and items to be recommended into the optimal multi-level information interaction model for processing, and output the preference score of each user for each item. Then, select the top H items with the highest predicted score for each user and which have not been interacted with to form each recommendation list and recommend them to the corresponding user.

[0102] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0103] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. A multi-layer information interaction recommendation method based on a gated graph neural network, characterized in that, Includes the following steps: Step 1: Collect a user set U = {u1, u2, ..., u...} from online social platforms. m ,…,u M }, Project set V = {v1, v2, ..., v n ,…,v N } and group collection Among them, u m Let v represent the m-th user. n This represents the nth item. Let G represent the k-th group, M represent the number of users, N represent the number of items, and K represent the number of groups. In the group set G, each group is a set consisting of a certain number of users and items. Step 2, construct three network diagrams, including: a user-project interaction diagram, a user social relationship diagram, and a project-related relationship diagram: Step 2.1: Construct a user-project interaction graph G based on the direct interaction relationship between users and projects. t (U∪V,e), where E represents the set of edges representing user-item interaction relationships. If the m-th user u m With the nth item v n If an interaction occurs, then the m-th user u m and the nth item v n There is an edge between them; Step 2.2: Construct a user social relationship graph G based on the social relationships between users in the user set U. u (U,S,X), where... Represents the m-th user u m and the c-th user u c Do they belong to the kth group? Let X represent the social relationships, and let X be the weighted neighbor matrix. The element X in the m-th row and c-th column of X is obtained by equation (1). mc : In equation (1), and Representing the m-th user u m The set of groups joined and the c-th user u c The group collection to join Step 2.3: Based on the relationships between the projects in the project set V, construct the project relationship graph G. v (V,C,Y), where... v represents the s-th item s and the tth project v t Whether or not they are shared in the kth group The correlation is given by Y, which represents the weighted neighbor matrix, and the element Y in the s-th row and t-th column of Y is obtained by equation (2). st : In equation (2), and These represent the s-th project v that has been shared. s The group set and the t-th item v t A collection of groups, Step 3: Construct a multi-layer gated graph neural network for information interaction, including: an embedding layer, an interaction graph information propagation layer, a relationship graph information propagation layer, a gating layer, and a prediction layer. Step 3.1, the embedding layer for the m-th user u m and the nth item v n Preprocessing is performed to obtain the m-th user u m Initial user embedding vector and the nth item v n Initial item embedding vector Step 3.2, the interactive graph information propagation layer respectively... and G t (U∪V,E) and and G t Process (U∪V,E) to obtain the m-th user u. m In G t User interaction information aggregation embedding in layer l of (U∪V,E) and the nth item v n In G t The project interaction information aggregation embedding in the l-th layer of (U∪V,E) Step 3.3, the relationship graph information propagation layer respectively... and G u (U,S,X) and and G v Process (V,C,Y) to obtain the m-th user u. m In G u User relationship information aggregation embedding in layer l of (U,S,X) and the nth item v n In the project relationship diagram G v Project relationship information aggregation embedding in the l-th layer of (V,C,Y) Step 3.4, the gating layer... and Perform fusion to obtain the m-th user u m The l-th layer user embedding vector and the nth item v n The l-th layer item embedding vector Step 3.5: Obtain the m-th user u following the process from Steps 3.2 to 3.

4. m Lth layer user embedding vector and the nth item v n The Lth layer of the project embedding vector layer Thus predicting the m-th user u m For the nth item v n Preference score Step 3.6: With the goal of minimizing the BPR loss function Loss, the parameters of the gated graph neural network with multi-layer information interaction are optimized and updated through the backpropagation algorithm and the Adam algorithm until the number of iterations reaches the set upper limit, thereby obtaining the optimal multi-layer information interaction model after training. Step 4: Input the new set of users and items to be recommended into the optimal multi-level information interaction model for processing, and output the preference score of each user for each item. Then, select the top H items with the highest predicted score for each user and which have not been interacted with to form each recommendation list and recommend them to the corresponding users.

2. The multi-layer information interaction recommendation method based on a gated graph neural network according to claim 1, characterized in that, Step 3.2 includes the following steps: Step 3.2.1, use equation (3) to calculate the m-th user u m Its neighboring project node v b The attention coefficient of the l-th layer In equation (3), It is the m-th user u m In G t Neighboring project nodes in (U∪V,E), N t (u m ) represents G t The m-th user u in (U∪V,E) m The set of neighboring project nodes, and Representing the m-th user u m and v b In G t User embedding vector and item embedding vector in layer (l-1) of (U∪V,E) Indicates in G t In (U∪V,E), the neighboring item node v at layer l is... b For the m-th user u m The importance of f, where f represents a single-layer feedforward neural network. G represents u In (U,S,X), the shared weight matrix of the l-th layer, [·||·] represents the concatenation operation; Step 3.2.2, use equation (4) to obtain G t In (U∪V,E), the neighboring item node v at layer l is... b For the m-th user u m Attention coefficient after regularization In equation (4), G represents t Another neighboring item node in layer l of (U∪V,E) For the m-th user u m Attention coefficient; Step 3.2.3, use equation (5) to obtain the m-th user u m In the user-project interaction graph G t User interaction information aggregation embedding in layer l of (U∪V,e) In equation (5), σ(·) represents the Sigmoid function. Represents the neighboring project node v b In G t The item embedding vector of the (l-1)th layer in (U∪V,E); Step 3.2.4: Following the process from steps 3.3.1 to 3.3.3, obtain the nth item v. n In the user-project interaction graph G t The project interaction information aggregation embedding in the l-th layer of (U∪V,E) 3. The multi-layer information interaction recommendation method based on a gated graph neural network according to claim 2, characterized in that, Step 3.3 includes the following steps: Step 3.3.1, use equation (6) to calculate the m-th user u m Its neighboring user node u c Normalized edge weights a mc : In equation (6), Represents the m-th user u m In the user social relationship graph G u Degree in (U,S,X) Represents the m-th user u m In the user social relationship graph G u The set of neighboring users in (U,S,X); Step 3.3.2, use equation (7) to obtain the m-th user u m In the user social relationship graph G u User relationship information aggregation embedding in layer l of (U,S,X) In equation (7), G represents u The shared weight matrix of the l-th layer in (U,S,X), where σ(·) represents the ReLU activation function, b l G represents u In (U,S,X), the bias parameters to be learned in the l-th layer are given, where L represents the optimal number of network layers. Represents the m-th user u m The (l-1)th layer user embedding vector; Step 3.3.3: Obtain the nth item v following the process of steps 3.2.1-3.2.

2. n In the project relationship diagram G v Project relationship information aggregation embedding in the l-th layer of (V,C,Y) 4. The multi-layer information interaction recommendation method based on a gated graph neural network according to claim 3, characterized in that, Step 3.4 involves using equations (8) and (9) to calculate... and In equations (8) and (9), ω1 and ω2 represent two vectors to be learned, and σ(·) represents the Sigmoid function. It represents the Hadamardi (or Hadama) stack.

5. The multi-layer information interaction recommendation method based on a gated graph neural network according to claim 4, characterized in that, In step 3.6, the BPR loss function Loss is constructed using equation (10): In equation (10), Represents the m-th user u m The sample set consisting of items in the item set V, n + Indicates the relationship with the m-th user u m Projects that have generated interaction, n - This indicates that a random sample is taken from the project set V and the m-th user u is selected. m Projects that have not generated any interaction This represents the set of positive samples consisting of user-project interactions. Let represent the set of negative user-item interactions randomly sampled from user set U and item set V; σ(·) represents the Sigmoid function, Θ represents all parameters to be trained in the model, and λ is the regularization coefficient.

6. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports the processor in executing any of the multi-layer information interaction recommendation methods of claims 1-5, and the processor is configured to execute the program stored in the memory.

7. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is run by the processor, it performs the steps of any of the multi-layer information interaction recommendation methods described in claims 1-5.

Images