CN117994007A - Social recommendation method based on multi-view fusion heterogeneous graph neural network - Google Patents
Social recommendation method based on multi-view fusion heterogeneous graph neural network Download PDFInfo
- Publication number
- CN117994007A CN117994007A CN202410398812.XA CN202410398812A CN117994007A CN 117994007 A CN117994007 A CN 117994007A CN 202410398812 A CN202410398812 A CN 202410398812A CN 117994007 A CN117994007 A CN 117994007A
- Authority
- CN
- China
- Prior art keywords
- node
- user
- view
- representation
- semantic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 230000004927 fusion Effects 0.000 title claims abstract description 29
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 18
- 239000011159 matrix material Substances 0.000 claims abstract description 47
- 238000010586 diagram Methods 0.000 claims abstract description 4
- 230000008569 process Effects 0.000 claims description 17
- 235000019580 granularity Nutrition 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 9
- 238000013507 mapping Methods 0.000 claims description 6
- 238000005096 rolling process Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 8
- 238000010168 coupling process Methods 0.000 description 4
- 238000005859 coupling reaction Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 230000004931 aggregating effect Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a social recommendation method based on a multi-view fusion heterogeneous graph neural network, which belongs to the technical field of social network recommendation and comprises the following steps: step 1, acquiring a relation between a user and a commodity in a social network, and establishing a heterogeneous diagram according to whether a link relation exists between the user and the commodity; step 2, an attribute topology decoupling module in the node view is established, and the established matrix is input into the attribute topology decoupling module to be embedded into the user under the node view; step 3, establishing a node coding module in the network mode view to obtain an embedded representation of a user in the network mode view; step 4, establishing a semantic fusion module in the semantic view to obtain an embedded representation of the user in the semantic view; and 5, carrying out fusion strategy of multi-view node representation, carrying out personalized recommendation on vector representation of user information. The method explores multi-granularity information in the social network from three perspectives, and more accurate commodity is recommended for users.
Description
Technical Field
The invention belongs to the technical field of social network recommendation, and particularly relates to a social recommendation method based on a multi-view fusion heterogeneous graph neural network.
Background
Social networks typically exhibit complex topologies, including the manner of connection between nodes, the degree distribution of nodes, and community structure. These structures may be sparse, dense, and there may be a high degree of bunching and small world nature. In a heterograph neural network, nodes may contain different types of attribute information, for example, in a social network, user nodes may contain age, gender, hobbies of interest, etc., and commodity nodes may contain price, category, sales, etc. Through the heterogeneous graph neural network, the various attribute information can be effectively combined for recommending tasks or other applications.
In the prior art, heterogeneous Graph Structure Learning (HGSL) fuses the attribute graph, the semantic graph and the original structure graph of the user together to generate a more comprehensive social network structure. ie-HGCN takes the graph data of the social network directly as input, performs a multi-layer convolution to learn the user node representation of the particular task. The SR-HGN is carefully designed for learning node representations in a social information network. It implements hierarchical learning by aggregating information at the user and type level, eliminating the need for a priori knowledge in semantic path selection. These prior art methods often result in poor classification and clustering effects for users in a social network because of the over-coupling of fine-grained information and the insufficient exploration of multi-grained information.
Disclosure of Invention
In order to solve the problems, the invention provides a social recommendation method based on a multi-view fusion heterogeneous graph neural network, which comprises the steps of designing an attribute-topology decoupling strategy, wherein the strategy converts the attribute of a user and the topology of the user respectively, so that fine granularity information of the user is protected, and transmission of multi-source information between adjacent nodes is avoided; meanwhile, a new social network structure learning strategy is designed, and subgraphs under different semantic relations are fused through graph-level attention, so that not only can the importance of different semantics be identified, but also interaction between a user and a commodity can be captured.
The technical scheme of the invention is as follows:
a social recommendation method based on a multi-view fusion heterogeneous graph neural network comprises the following steps:
Step 1, acquiring a relation between a user and a commodity in a social network, and establishing a heterogeneous diagram according to whether a link relation exists between the user and the commodity;
Step 2, an attribute topology decoupling module in the node view is established, and the established matrix is input into the attribute topology decoupling module to perform feature and topology decoupling coding to obtain node embedded representation in the node view;
step 3, a node coding module in the network mode view is established, and node coding is carried out to obtain node embedded representation in the network mode view;
step 4, establishing a semantic fusion module in the semantic view, and carrying out semantic fusion to obtain node embedded representation in the semantic view;
and 5, carrying out fusion strategies of multi-view node representation, fusing information of different granularities of the social network by utilizing the splicing strategies, obtaining vector representations of user information, and carrying out personalized recommendation.
Further, the specific process of the step 1 is as follows: matrix for built heterogeneous graphRepresentation of/>，Representing the total number of user nodes and corresponding the total number of rows of the matrix; /(I)Representing the total number of commodity nodes and the total number of columns of the corresponding matrix; the rows of the matrix represent user nodes and the columns represent commodity nodes, if/>Individual user nodes and/>Links exist between individual commodity nodes, then the matrix's/>Line/>The column value is set to 1, otherwise to 0.
Further, the specific process of the step 2 is as follows:
Step 2.1, matrix is formed Input attribute topology decoupling module, for one type/>User node/>By mapping matrix/>User node/>Attribute characteristics/>Projection to a common dimension interval is as follows:
(1)；
Wherein, Is the user node/>Projection features of/>Is a dimension; /(I)Is an activation function; /(I)Representing vector bias;
Step 2.2, user node Projection features/>And topology/>Coding to a common dimension interval through a multi-layer perceptron, wherein the method comprises the following steps of:
(2)；
(3)；
Wherein, Is the encoded user node/>Attribute information representation of (a); /(I)Is the encoded user node/>Is represented by topology information of (a); /(I)Is a multi-layer perceptron;
Step 2.3, splicing the two coded representations, mapping the attribute information of the user and the commodity to a shared dimension through a multi-layer perceptron and an activation function, and finally obtaining the user node under the node view Embedded representation/>The method is characterized by comprising the following steps:
(4)；
Wherein, Representing a stitching operation.
Further, the specific process of the step 3 is as follows:
User node With other types of nodes/>Connected,/>、/>、、/>Representing different node types;
Applying node level attention to Type of surrounding neighbor:
(5)；
Wherein, Is the user node/>The type of connection is/>Is embedded with a representation of a neighbor node; /(I)For user node/>The type of connection is/>Is a neighbor node of (a); /(I)Is a neighbor node/>Is a projection feature of (2);
Is of the type/> Neighbor node/>For user node/>Is calculated as follows:
(6)；
Wherein, Is an exponential function based on a natural constant e; /(I)Is an activation function; Is of the type/> Is a node level attention vector of (1); /(I)Transpose the symbol; /(I)For user node/>Information of (2); /(I)The symbol is spliced;
Obtaining the user node according to the formula (5) Embedding representations for all different types of neighbor nodes of a connection；
Fusing all different types of neighbor node embedded representations together with type-level attention to obtain user nodes in network mode viewEmbedded representation/>; The formula is as follows:
(7)；
Wherein, For the total number of types,/>Is a type index.
Further, the specific process of the step 4 is as follows:
Step 4.1, dividing an original social network into different graph structures through semantic knowledge, and decomposing modeled graph data into different sub-modules to learn different semantic information; different semantics are fused through the graph-level attention as an integrated social network structure, and the method is calculated as follows:
(8)；
Wherein, Combining the different semantic relations to obtain a final relation matrix; /(I)Representation with parameters/>For assigning different weights to each switching matrix; /(I)For semantic relation/>Is a matrix of (a); /(I)For semantic relation/>Is a matrix of (a); /(I)For semantic relation/>Is a matrix of (a);
Step 4.2, will The node representation is obtained by being put into a graph rolling network GCN, and is calculated as follows:
(9)；
Wherein, For user node/>, under semantic viewIs embedded in the representation; /(I)Is a normalized relationship matrix; /(I)Is a user feature; /(I)Is a weight parameter.
Further, the specific process of the step 5 is as follows:
Splicing the obtained embedded representations of the user nodes in the three views together, and obtaining the final embedded representation of the nodes through a multi-layer perceptron:
(10)；
Wherein, For user node/>The embedded representation is a learned representation of the user information, and then personalized recommendations or predictions of user preferences are made based on the user's behavior and attribute information.
The invention has the beneficial technical effects that: the feature and the topology information of the user are processed separately, so that the excessive coupling of fine granularity information caused by the information transmission of multi-source heterogeneous information can be decoupled, the independent modeling of the attribute information and the topology information of the user is realized, the excessive coupling of the fine granularity information of the user is decoupled, and the technology is applied to a heterogeneous graph neural network model, so that the classification and clustering performance of the nodes can be improved; three views are designed to explore multi-granularity information in a social network, the extraction of fine granularity information of a user is realized through a node view, the extraction of topology information of neighbors around the user is realized through a network mode view, the capture of high-order heterogeneous information of the user is realized through a semantic view, and then the exploration of multi-granularity information in the social network is realized. According to the invention, the heterogeneous graph neural network is utilized to simultaneously learn a simple topological structure and a complex semantic relation between the user and the commodity, and different information is fused through a splicing strategy, so that the accuracy of user classification and clustering is improved, and more accurate commodity is recommended for the user.
Drawings
FIG. 1 is a flow chart of a social recommendation method based on a multi-view fusion heterogeneous graph neural network.
FIG. 2 is a diagram showing the complexity of the model of the present invention compared with other models.
Detailed Description
The invention is described in further detail below with reference to the attached drawings and detailed description:
The invention takes a social network as a research object, improves classification and clustering of users in the social network as a core target, and the key technical problems to be solved include: first: the attribute information of the user is coupled. Second,: multi-granularity information of social networks is not fully mined. Solving both of these issues can accomplish modeling of complex social networks.
Therefore, the specific key problems to be solved by the invention are as follows:
Key technical problem 1: attribute information of the user is coupled;
The invention designs a node view angle and a network mode view angle to decouple entanglement between users and commodities caused by isomerism. Specifically, the node views model the user's properties and decoupling information, respectively, to obtain fine-grained information of the user. The network mode view obtains local topology information of the user by modeling commodity attributes and structure information of a neighborhood around the user.
Key technical problem 2: multi-granularity information in an undermined social network;
The invention designs three views of a node view, a network mode view and a semantic view to learn different semantic information, and digs different granularity information of users in the social network. The node view decouples semantic structures of the learning user and the commodity respectively; the network mode visual angle learns the network mode between the user and the commodity, and the semantic visual angle learns the complex semantics of the user and the commodity. Through the three views, attribute information and topological structure information of users in the social network and high-order heterostructure information are captured, and multi-granularity information mining in the social network is realized.
As shown in fig. 1, the invention provides a social recommendation method of a heterogeneous graph neural network based on multi-view fusion, which comprises the following steps:
Step 1, acquiring a relation between a user and a commodity in a social network, and establishing a heterogeneous graph according to whether a link relation exists between the user and the commodity. The specific process is as follows: a heterogram is a network that represents a complex network of relationships. A heterogram is an abstract representation that requires modeling of network data into a matrix vector form to transform the abstract into concrete to apply knowledge learning. Thus, the matrix for heterogeneous image established by the invention Representation of/>，/>Representing the total number of user nodes and corresponding the total number of rows of the matrix; /(I)And the total number of commodity nodes is represented, and the total column number of the matrix is correspondingly represented. The rows of the matrix represent user nodes and the columns represent commodity nodes, if/>Individual user nodes and/>Links exist between individual commodity nodes, then the matrix's/>Line 1The column value is set to 1, otherwise to 0.
And 2, establishing an attribute topology decoupling module in the node view, inputting the established matrix into the attribute topology decoupling module, and performing feature and topology decoupling coding to obtain node embedded representation in the node view. The node views model the user attributes and topology, respectively, to obtain fine-grained information. To avoid over-coupling of the attributes and topology of the user, heterogeneous messaging is split into two modules for separate processing. In the attribute-topology decoupling module, only the attribute of the user and the adjacent topology are respectively transformed. Thus, only fine granularity information of the user is reserved, and transmission of multi-source heterogeneous information from adjacent structures is avoided. The specific process is as follows:
Step 2.1, matrix is formed And inputting an attribute topology decoupling module. Because there are different types of user nodes in a social network, the target features of user nodes are typically located in unused space, and therefore, it is first necessary to project all types of user node features into a common potential vector space, specifically/>, for one typeUser node/>By a mapping matrix of a specific type/>User node/>Attribute characteristics/>Projection to a common dimension interval is as follows:
(1)；
Wherein, Is the user node/>Projection features of/>Is a dimension; /(I)Is an activation function; /(I)Representing the vector bias.
Attribute embedding can combine the attribute information and topology information of the user nodes together to obtain a more comprehensive representation of the user nodes, which can make the user node embedding more accurate and reliable.
Step 2.2, user nodeProjection features/>And topology/>Encoding to a common dimension interval through a multi-layer perceptron (MLP), specifically as follows:
(2)；
(3)；
Wherein, Is the encoded user node/>Attribute information representation of (a); /(I)Is the encoded user node/>Is represented by topology information of (a); /(I)Is a multi-layer perceptron.
Step 2.3, splicing the two coded representations, mapping the attribute information of the user and the commodity to a shared dimension through a multi-layer perceptron and an activation function, and finally obtaining the user node under the node viewEmbedded representation/>The method is characterized by comprising the following steps:
(4)；
Wherein, Representing a stitching operation.
And 3, establishing a node coding module in the network mode view, and performing node coding to obtain node embedded representation in the network mode view. The network mode view obtains local information of the user by modeling attribute and structure information of a neighborhood around the user. In the node coding module of the network mode, the information of the user is aggregated by adopting a two-stage attention mechanism. The specific process is as follows:
the goal of the node encoding module is to learn the user node embedding in network mode. Suppose a user node With other types of nodes/>Connected,/>、/>、/>、/>Representing different node types,/>For the total number of types,/>For type index, then user node/>The type of connection is/>May be expressed as a neighbor node of (a). For user node/>The different types of neighbors contribute differently to the representation of the user node, so that the contribution of the same type of neighbor node is different, and therefore, the node-level and type-level attention mechanisms are adopted to aggregate the other types of neighbors to the user node/>, in a hierarchical mannerIs a message of (a).
Specifically, node level attention is first applied toType of surrounding neighbor:
(5)；
Wherein, Is the user node/>The type of connection is/>Is embedded with a representation of a neighbor node; /(I)For user node/>The type of connection is/>Is a neighbor node of (a); /(I)Is a neighbor node/>Is a projection feature of (2);
Is of the type/> Neighbor node/>For user node/>Is calculated as follows:
(6)；
Wherein, Is an exponential function based on a natural constant e; /(I)Is an activation function; Is of the type/> Is a node level attention vector of (1); /(I)Transpose the symbol; /(I)For user node/>Information of (2); /(I)For splice symbols.
In practical use, the non-convergence comes fromInstead, part of surrounding neighbor information is randomly extracted in each epoch to be aggregated, and one epoch represents the update times when all training data is used once in learning. In particular, if the type is/>The number of surrounding neighbors exceeds a preset threshold/>Randomly selecting neighbor type as/>As/>Without selecting all neighbor nodes; if the number of neighbor nodes is less than the threshold/>The selection type is repeated as/>Until reaching the threshold. In this way, each node is guaranteed to aggregate the same amount of information from neighbors, and the diversity of node embeddings is increased in each epoch.
Obtaining the user node according to the formula (5)Embedding representations for all different types of neighbor nodes of a connection；
Fusing them together using mean fusion to obtain user nodes in network mode viewIs embedded in the representation of (a); The formula is as follows:
(7)；
The hierarchical attention mechanism can distinguish different types of nodes and the contributions of the same type of nodes, and fine-grained representation learning is realized.
And 4, establishing a semantic fusion module in the semantic view, and carrying out semantic fusion to obtain embedded representation in the semantic view. Semantic views understand the high-level heterogeneous information of a user by merging different sub-graphs into a single graph through graph-level focus. In order to mine multi-granularity information, semantic knowledge is used for aggregating high-order information in a semantic fusion module, different semantic knowledge is fused through channel attention, and semantic interaction is achieved. The specific process is as follows:
Step 4.1, dividing an original social network into different graph structures through semantic knowledge, and decomposing modeled graph data into different sub-modules to learn different semantic information; they are fused by graph level attention as an integrated social network structure, calculated as follows:
(8)；
Wherein, Combining the different semantic relations to obtain a final relation matrix; /(I)Representation with parameters/>For assigning different weights to each switching matrix; /(I)For semantic relation/>Is a matrix of (a); /(I)For semantic relation/>Is a matrix of (a); /(I)For semantic relation/>Is a matrix of (a) in the matrix.
Step 4.2, willFetching node representations (in fact,/>) into a graph rolling network GCNCan be applied to any graph neural network GNN encoder), and is calculated as follows:
(9)；
Wherein, For user node/>, under semantic viewIs embedded in the representation; /(I)Is a normalized relationship matrix; /(I)Is a user feature; /(I)Is a weight parameter.
And 5, carrying out fusion strategies of multi-view node representation, and fusing different granularity information of the social network by utilizing the splicing strategies to obtain vector representations of user information, so as to carry out personalized recommendation. The specific process is as follows:
The obtained embedded representations of the user nodes under the three views are effectively spliced together, and the final embedded representation of the nodes is obtained through a multi-layer perceptron:
(10)；
Wherein, For user node/>The embedded representation is a learned representation of the user information, and then personalized recommendations or predictions of user preferences are made based on the user's behavior and attribute information.
In order to demonstrate the feasibility and superiority of the present invention, the following comparative experiments were performed.
Experiment 1: the above-mentioned attribute topology decoupling module, node encoding module, semantic fusion module, and fusion strategy of multi-view node representation essentially constitute the model MHGNN of the present invention. The experimental data set of the invention comprises ACM, IMDB, DBLP, YELP data sets. Four segmentation rates of 20%, 40%, 60%, 80% were used for each dataset. First, the results of seven baseline models MHGNN and HAN, GTN, MAGNN, HGSL, HPN, roHe, ie-HGCN in semi-supervised node classification are evaluated, macro-F1 and Micro-F1 are selected by an evaluation index method, and specific data results are shown in table 1. Comparing the model with seven representative algorithm models, wherein Macro-F1 and Micro-F1 are two calculation modes of F1-score, and F1-score is an index for measuring the accuracy of the two classification models and is used for measuring the accuracy of unbalanced data. The higher the Macro-F1 or Micro-F1 value, the higher the model accuracy. The HAN is divided into different semantic subgraphs through different semantic relations between the user and the commodity, and then the different subgraphs are fused with the user characterization under the different subgraphs through attention. The GTN iteratively convolves the user information through a relation matrix between the user and the commodity. MAGNN by coding the relationship between the user and the commodity into a sequence form, all node attribute information is reserved, but in the coding process, the user information is repeatedly coded for a plurality of times, so that information redundancy is caused. HGSL by using user attributes and neighborhood structure, learn the correct link relationships, eliminate noise information present in the original graph topology. HPN alleviates the problem of semantic confusion after multi-layer convolution by preserving the initial relationship matrix at each layer. RoHe, by way of a form of attention-directed edge, works anyway efficiently in the case of noise attacks. ie-HGCN iteratively couples the information of the surrounding neighborhood layer by relational convolution. The invention processes the social network data into a proper format, feeds the format back to the model, and tests the superiority of the method proposed by us.
Table 1 results (%) of each model on semi-supervised node classification;
。
As can be seen from Table 1, the model MHGNN of the present invention performs well on the ACM, IMDB, DBLP, YELP four datasets. In general, the model MHGNN of the present invention, by aggregating information of different granularity at three perspectives, yields a better representation of the user than a representation of the user at a single perspective by semantic path or modeling the user's own information, because the MHGNN model not only captures high-order heterogeneous information using semantic perspectives, but also enables a stronger representation of the learned user by network mode perspectives and by decoupling features and topologies.
Experiment 2: the present invention also performed a node clustering experiment to verify the performance of the model MHGNN of the present invention. Specifically, the obtained user representation is put into K-means for clustering, and the clustering number is set as the true class number of the node. Results of two evaluation index methods, average Normalized Mutual Information (NMI) and Adjusted Rand Index (ARI), are reported. The higher the NMI or ARI value, the better the model performance.
Table 2 MHGNN results (%) on node clusters;
。
As can be seen from table 2 MHGNN performed better than the other baseline models. Compared with other models with single view angles, MHGNN fully grabs the semantic relation between the nodes and the surrounding neighborhood, so that the degree of distinction between the user and the neighborhood nodes is higher, and the clustering effect is more obvious.
FIG. 2 compares and analyzes the complexity of the six models MHGNN and HAN, MAGNN, HGSL, HPN, roHe, ie-HGCN of the present invention. As can be seen from FIG. 2, although MHGNN uses three views to learn the user's representation, it does not consume much memory and time, because the three views eventually use stitching operation, and the attention is used to merge the sub-graphs under different semantic paths into one meta-graph, so that the time for computing each sub-graph is saved, the computation amount is reduced, and the computation efficiency is improved.
It should be understood that the above description is not intended to limit the invention to the particular embodiments disclosed, but to limit the invention to the particular embodiments disclosed, and that the invention is not limited to the particular embodiments disclosed, but is intended to cover modifications, adaptations, additions and alternatives falling within the spirit and scope of the invention.
Claims (6)
1. The social recommendation method based on the multi-view fusion heterogeneous graph neural network is characterized by comprising the following steps of:
Step 1, acquiring a relation between a user and a commodity in a social network, and establishing a heterogeneous diagram according to whether a link relation exists between the user and the commodity;
Step 2, an attribute topology decoupling module in the node view is established, and the established matrix is input into the attribute topology decoupling module to perform feature and topology decoupling coding to obtain node embedded representation in the node view;
step 3, a node coding module in the network mode view is established, and node coding is carried out to obtain node embedded representation in the network mode view;
step 4, establishing a semantic fusion module in the semantic view, and carrying out semantic fusion to obtain node embedded representation in the semantic view;
and 5, carrying out fusion strategies of multi-view node representation, fusing information of different granularities of the social network by utilizing the splicing strategies, obtaining vector representations of user information, and carrying out personalized recommendation.
2. The social recommendation method based on the multi-view fusion heterogeneous graph neural network according to claim 1, wherein the specific process of the step 1 is as follows: matrix for built heterogeneous graphRepresentation of/>，/>Representing the total number of user nodes and corresponding the total number of rows of the matrix; /(I)Representing the total number of commodity nodes and the total number of columns of the corresponding matrix; the rows of the matrix represent user nodes and the columns represent commodity nodes, if/>Individual user nodes and/>Links exist between individual commodity nodes, then the matrix's/>Line 1The column value is set to 1, otherwise to 0.
3. The social recommendation method based on the multi-view fusion heterogeneous graph neural network according to claim 2, wherein the specific process of the step 2 is as follows:
Step 2.1, matrix is formed Input attribute topology decoupling module, for one type/>User node/>By mapping matrix/>User node/>Attribute characteristics/>Projection to a common dimension interval is as follows:
(1)；
Wherein, Is the user node/>Projection features of/>Is a dimension; /(I)Is an activation function; /(I)Representing vector bias;
Step 2.2, user node Projection features/>And topology/>Coding to a common dimension interval through a multi-layer perceptron, wherein the method comprises the following steps of:
(2)；
(3)；
Wherein, Is the encoded user node/>Attribute information representation of (a); /(I)Is the encoded user node/>Is represented by topology information of (a); /(I)Is a multi-layer perceptron;
Step 2.3, splicing the two coded representations, mapping the attribute information of the user and the commodity to a shared dimension through a multi-layer perceptron and an activation function, and finally obtaining the user node under the node view Embedded representation/>The method is characterized by comprising the following steps:
(4)；
Wherein, Representing a stitching operation.
4. The social recommendation method based on the multi-view fusion heterogeneous graph neural network according to claim 3, wherein the specific process of the step 3 is as follows:
User node With other types of nodes/>Connected,/>、/>、/>、Representing different node types;
Applying node level attention to Type of surrounding neighbor:
(5)；
Wherein, Is the user node/>The type of connection is/>Is embedded with a representation of a neighbor node; /(I)For user node/>The type of connection is/>Is a neighbor node of (a); /(I)Is a neighbor node/>Is a projection feature of (2);
Is of the type/> Neighbor node/>For user node/>Is calculated as follows:
(6)；
Wherein, Is an exponential function based on a natural constant e; /(I)Is an activation function; Is of the type/> Is a node level attention vector of (1); /(I)Transpose the symbol; /(I)For user node/>Information of (2); /(I)The symbol is spliced;
Obtaining the user node according to the formula (5) Embedding representations for all different types of neighbor nodes of a connection；
Fusing all different types of neighbor node embedded representations together with type-level attention to obtain user nodes in network mode viewEmbedded representation/>; The formula is as follows:
(7)；
Wherein, For the total number of types,/>Is a type index.
5. The social recommendation method based on the multi-view fusion heterogeneous graph neural network according to claim 4, wherein the specific process of the step 4 is as follows:
Step 4.1, dividing an original social network into different graph structures through semantic knowledge, and decomposing modeled graph data into different sub-modules to learn different semantic information; different semantics are fused through the graph-level attention as an integrated social network structure, and the method is calculated as follows:
(8)；
Wherein, Combining the different semantic relations to obtain a final relation matrix; /(I)Representation with parameters/>For assigning different weights to each switching matrix; /(I)For semantic relation/>Is a matrix of (a); /(I)Is a semantic relationshipIs a matrix of (a); /(I)For semantic relation/>Is a matrix of (a);
Step 4.2, will The node representation is obtained by being put into a graph rolling network GCN, and is calculated as follows:
(9)；
Wherein, For user node/>, under semantic viewIs embedded in the representation; /(I)Is a normalized relationship matrix; /(I)Is a user feature; /(I)Is a weight parameter.
6. The social recommendation method based on the multi-view fusion heterogeneous graph neural network according to claim 5, wherein the specific process of step5 is as follows:
Splicing the obtained embedded representations of the user nodes in the three views together, and obtaining the final embedded representation of the nodes through a multi-layer perceptron:
(10)；
Wherein, For user node/>The embedded representation is a learned representation of the user information, and then personalized recommendations or predictions of user preferences are made based on the user's behavior and attribute information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410398812.XA CN117994007B (en) | 2024-04-03 | 2024-04-03 | Social recommendation method based on multi-view fusion heterogeneous graph neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410398812.XA CN117994007B (en) | 2024-04-03 | 2024-04-03 | Social recommendation method based on multi-view fusion heterogeneous graph neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117994007A true CN117994007A (en) | 2024-05-07 |
| CN117994007B CN117994007B (en) | 2024-07-05 |
Family
ID=90896347
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410398812.XA Active CN117994007B (en) | 2024-04-03 | 2024-04-03 | Social recommendation method based on multi-view fusion heterogeneous graph neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117994007B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104090957A (en) * | 2014-03-10 | 2014-10-08 | 中国科学院软件研究所 | Heterogeneous network interactive visualization method |
| CN108491680A (en) * | 2018-03-07 | 2018-09-04 | 安庆师范大学 | Drug relationship abstracting method based on residual error network and attention mechanism |
| CN108804677A (en) * | 2018-06-12 | 2018-11-13 | 合肥工业大学 | In conjunction with the deep learning question classification method and system of multi-layer attention mechanism |
| CN109754317A (en) * | 2019-01-10 | 2019-05-14 | 山东大学 | Merge interpretation clothes recommended method, system, equipment and the medium of comment |
| CN109862585A (en) * | 2019-01-31 | 2019-06-07 | 湖北工业大学 | A kind of dynamic heterogeneous network flow prediction method based on depth Space-time Neural Network |
| CN111400560A (en) * | 2020-03-10 | 2020-07-10 | 支付宝(杭州)信息技术有限公司 | Method and system for predicting based on heterogeneous graph neural network model |
| US20210118073A1 (en) * | 2019-10-22 | 2021-04-22 | International Business Machines Corporation | Land use planning recommendations using heterogeneous temporal datasets |
| CN113254803A (en) * | 2021-06-24 | 2021-08-13 | 暨南大学 | Social recommendation method based on multi-feature heterogeneous graph neural network |
| CN114090902A (en) * | 2021-11-22 | 2022-02-25 | 中国人民解放军国防科技大学 | Social network influence prediction method and device based on heterogeneous network |
| CN115082142A (en) * | 2022-05-10 | 2022-09-20 | 华南理工大学 | Recommendation method, device and medium based on heterogeneous relational graph neural network |
| CN115713386A (en) * | 2022-11-24 | 2023-02-24 | 山东大学 | Multi-source information fusion commodity recommendation method and system |
| CN115718826A (en) * | 2022-11-29 | 2023-02-28 | 中国科学技术大学 | Method, system, device and medium for classifying target nodes in graph structure data |
-
2024
- 2024-04-03 CN CN202410398812.XA patent/CN117994007B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104090957A (en) * | 2014-03-10 | 2014-10-08 | 中国科学院软件研究所 | Heterogeneous network interactive visualization method |
| CN108491680A (en) * | 2018-03-07 | 2018-09-04 | 安庆师范大学 | Drug relationship abstracting method based on residual error network and attention mechanism |
| CN108804677A (en) * | 2018-06-12 | 2018-11-13 | 合肥工业大学 | In conjunction with the deep learning question classification method and system of multi-layer attention mechanism |
| CN109754317A (en) * | 2019-01-10 | 2019-05-14 | 山东大学 | Merge interpretation clothes recommended method, system, equipment and the medium of comment |
| CN109862585A (en) * | 2019-01-31 | 2019-06-07 | 湖北工业大学 | A kind of dynamic heterogeneous network flow prediction method based on depth Space-time Neural Network |
| US20210118073A1 (en) * | 2019-10-22 | 2021-04-22 | International Business Machines Corporation | Land use planning recommendations using heterogeneous temporal datasets |
| CN111400560A (en) * | 2020-03-10 | 2020-07-10 | 支付宝(杭州)信息技术有限公司 | Method and system for predicting based on heterogeneous graph neural network model |
| CN113254803A (en) * | 2021-06-24 | 2021-08-13 | 暨南大学 | Social recommendation method based on multi-feature heterogeneous graph neural network |
| CN114090902A (en) * | 2021-11-22 | 2022-02-25 | 中国人民解放军国防科技大学 | Social network influence prediction method and device based on heterogeneous network |
| CN115082142A (en) * | 2022-05-10 | 2022-09-20 | 华南理工大学 | Recommendation method, device and medium based on heterogeneous relational graph neural network |
| CN115713386A (en) * | 2022-11-24 | 2023-02-24 | 山东大学 | Multi-source information fusion commodity recommendation method and system |
| CN115718826A (en) * | 2022-11-29 | 2023-02-28 | 中国科学技术大学 | Method, system, device and medium for classifying target nodes in graph structure data |
Non-Patent Citations (4)
| Title |
|---|
| CHAOLI等: "HetReGAT-FC: Heterogeneous Residual Graph Attention Network via Feature Completion", 《ELSEVIER》, 9 March 2023 (2023-03-09) * |
| ZHONGYING ZHAO等: "GuessUNeed: Recommending Courses via Neural Attention Network and Course Prerequisite Relation Embeddings", 《ACM》, 31 December 2020 (2020-12-31) * |
| 贾霄生等: "互信息与多条元路径融合的异质网络表示学习方法", 《软件学报》, 31 December 2023 (2023-12-31) * |
| 高莉: "基于图神经网络的社交推荐系统研究", 《信息科技》, 15 February 2023 (2023-02-15) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117994007B (en) | 2024-07-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112529168B (en) | GCN-based attribute multilayer network representation learning method | |
| Liu et al. | Graph summarization methods and applications: A survey | |
| Zhu et al. | A survey on graph structure learning: Progress and opportunities | |
| CN111881350B (en) | Recommendation method and system based on mixed graph structured modeling | |
| Rhee et al. | Active and semi-supervised learning for object detection with imperfect data | |
| Yang et al. | From properties to links: Deep network embedding on incomplete graphs | |
| CN113918832B (en) | Graph convolution collaborative filtering recommendation system based on social relationship | |
| CN114637923B (en) | Data information recommendation method and device based on hierarchical attention-graph neural network | |
| CN113918833B (en) | Product recommendation method realized through graph convolution collaborative filtering of social network relationship | |
| Cai et al. | Graph domain adaptation: A generative view | |
| CN113268993B (en) | Mutual information-based non-supervision network representation learning method for attribute heterogeneous information network | |
| CN113918834B (en) | Graph convolution collaborative filtering recommendation method fusing social relations | |
| Wang et al. | Dynamic dense graph convolutional network for skeleton-based human motion prediction | |
| Sun et al. | Attention-based graph neural networks: a survey | |
| CN115221413B (en) | Sequence recommendation method and system based on interactive graph attention network | |
| Chu et al. | Variational cross-network embedding for anonymized user identity linkage | |
| Tan et al. | Collaborative graph neural networks for attributed network embedding | |
| Hao et al. | Deep graph clustering with enhanced feature representations for community detection | |
| Zhang et al. | End‐to‐end generation of structural topology for complex architectural layouts with graph neural networks | |
| Zhang et al. | A knowledge graph recommendation model via high-order feature interaction and intent decomposition | |
| Li et al. | Higher order heterogeneous graph neural network based on node attribute enhancement | |
| Peng et al. | Unsupervised Multiplex Graph Learning with Complementary and Consistent Information | |
| Zhao et al. | A novel embedding learning framework for relation completion and recommendation based on graph neural network and multi-task learning | |
| CN117994007B (en) | Social recommendation method based on multi-view fusion heterogeneous graph neural network | |
| CN116756432A (en) | Social user implicit relation recommendation method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |