A multi-view social network community discovery method, system and device

Abstract

The application discloses a multi-view social network community discovery method, system and device, relates to the field of social network data analysis, and comprises the following steps: acquiring multi-view network data; filtering and processing the multi-view network data by using a low-pass filter to obtain a smooth representation of each view; fusing the smooth representation of each view by using an attention mechanism fusion mechanism to obtain a fused representation; fusing a Laplace matrix corresponding to each view by using a learning weight to obtain a fused Laplace matrix; determining an embedding representation by using an MLP network based on the attention mechanism fusion mechanism which has been trained according to the fused representation and the fused Laplace matrix; and clustering the embedding representation to obtain a final clustering result, thereby realizing community discovery in a social network. The application can realize stable and accurate social network user community identification.

Description

Technical Field

[0001] This invention relates to the field of social network data analysis, and in particular to a method, system, and device for discovering communities in a multi-view social network. Background Technology

[0002] In recent years, with the rapid development of social network technologies, people's social lives have changed accordingly, bringing users entirely new experiences such as social network search, online shopping, and community services. At the same time, with the improvement of the informatization level of social networks, the scale of user data has continued to expand, and its quality has also improved. Relationships between users have become more diversified, including mutual friends, shared interests, and various interactive behaviors. This large-scale, high-quality, multi-view user data provides a more comprehensive perspective and richer data resources for social network management, user behavior analysis, and commercial marketing. Viewing social networks as an ecosystem, understanding users' interests, habits, and behaviors is of great significance for the development of social networks and the optimization of user experience.

[0003] In the era of big data, extracting knowledge from massive data resources has become a crucial research topic. Multi-view graph clustering is an unsupervised learning-based method designed to discover and reveal the inherent structure and patterns of social network datasets with multiple views. Multi-view graph clustering can be used for community discovery. In social networks, multi-view community discovery technology has been widely applied in areas such as customer relationship management, community discovery, user behavior analysis, and personalized recommendations. Community discovery attempts to divide samples from social network user datasets into several mutual communities to identify and classify user characteristics, thereby assisting in social network management and commercial marketing.

[0004] The rapid development of data mining technology has brought new opportunities and challenges to the field of social networks. User data on social networks is a typical type of multi-view network data, including users' basic attribute information, social relationships, interests, and behavioral records, which is a valuable social resource. Previously, social network data was mainly used to realize functions such as recommendation and advertising. With the rapid development of multi-view community discovery technology, users can be clustered based on user data and behavioral records between users on social networks to discover specific user groups, providing data support and assistance for subsequent social network analysis and recommendation. First, the data is preprocessed, and unstructured data is structured and used as input. Then, effective multi-view graph clustering analysis is performed on the social network data to achieve user group identification and interest group discovery. Since most existing technologies fail to learn corresponding weights based on the quality of multi-views, poor-quality multi-views may affect the final community discovery results.

[0005] Therefore, for multi-view network data, there is an urgent need to provide a robust and practical method or system for discovering multi-view social network communities, which can achieve stable and accurate identification of social network user groups. Summary of the Invention

[0006] The purpose of this invention is to provide a method, system, and device for discovering communities in multi-view social networks, which can achieve stable and accurate identification of social network user groups.

[0007] To achieve the above objectives, the present invention provides the following solution:

[0008] A method for discovering communities on a multi-view social network includes:

[0009] Acquire multi-view network data; the multi-view network data includes: basic attribute information, social relationships, relationships based on shared interests, and records of interactive behaviors;

[0010] The multi-view network data is filtered using a low-pass filter to obtain a smooth representation of each view;

[0011] The smooth representations of each view are fused using an attention-based fusion mechanism to obtain a fused representation;

[0012] The Laplacian matrix corresponding to each view is fused using the learned weights to obtain the fused Laplacian matrix;

[0013] Based on the fused representation and the fused Laplacian matrix, a trained MLP network based on the attention mechanism fusion mechanism is used to determine the embedding representation;

[0014] Clustering the embedding representation yields the final clustering result, enabling community discovery in social networks.

[0015] Optionally, the acquisition of multi-view network data further includes:

[0016] Preprocessing of multi-view network data includes filling in missing values ​​and removing outliers.

[0017] The preprocessed basic attribute information is represented as an X matrix.

[0018] Social relationships, shared interests, and interaction records are represented as an adjacency matrix A1, A2, ..., A1. M .

[0019] Optionally, the process of fusing the smooth representation of network data using an attention mechanism to obtain a fused representation specifically includes:

[0020] The initial fusion representation is determined based on the initial weights;

[0021] Spectral clustering is performed on the initial fusion representation to obtain the clustering result P;

[0022] Performing spectral clustering on the smooth representation yields the clustering result Q. m ;

[0023] Based on clustering results P and Q m Update the initial weights to obtain the current updated weights;

[0024] When the difference between the currently updated weight and the previously updated weight is less than the difference threshold, the fused representation is determined using the currently updated weight;

[0025] When the difference between the currently updated weight and the previously updated weight is not less than the difference threshold, return to the step of determining the initial fusion representation based on the initial weight, until the difference between the currently updated weight and the previously updated weight is less than the difference threshold.

[0026] Optionally, the step of fusing the Laplacian matrix corresponding to each view using learned weights to obtain a fused Laplacian matrix specifically includes:

[0027] Using formula Determine the fusion Laplacian matrix;

[0028] in, Let t = {t1, t2, ..., t} be the fused Laplace matrix. M} is the learning matrix, t m L represents the learning weights corresponding to the m-th view. m Let be the Laplacian matrix corresponding to the m-th view, where M is the number of views in the multi-view network data.

[0029] Optionally, the step of determining the embedding representation using a trained MLP network based on an attention-based fusion mechanism, according to the fused representation and the fused Laplacian matrix, specifically includes:

[0030] The fused representation and the fused Laplacian matrix are fused;

[0031] The fused data is then filtered to obtain the final representation H. α ;

[0032] According to the final representation H α We use a trained MLP network based on an attention-based fusion mechanism to determine the embedding representation Z.

[0033] Optionally, the training process of the MLP network based on the attention mechanism fusion mechanism is as follows:

[0034] Construct the training set and loss function based on the embedding representation Z;

[0035] Update the parameters and learned weights of the MLP network using the Adam optimizer.

[0036] A multi-view social network community discovery system, comprising:

[0037] A multi-view network data acquisition module is used to acquire multi-view network data; the multi-view network data includes: basic attribute information, social relationships, shared interests and hobbies, and interaction behavior records;

[0038] The smooth representation determination module is used to filter the multi-view network data using a low-pass filter to obtain a smooth representation of each view.

[0039] The fusion representation determination module is used to fuse the smooth representations of each view using an attention mechanism to obtain a fusion representation;

[0040] The Laplacian matrix fusion module is used to fuse the Laplacian matrix corresponding to each view using learned weights to obtain a fused Laplacian matrix.

[0041] The embedding representation determination module is used to determine the embedding representation based on the fused representation and the fused Laplacian matrix, using a trained MLP network based on the attention mechanism fusion mechanism.

[0042] The community detection module is used to cluster the embedded representations to obtain the final clustering results, thereby realizing community detection in social networks.

[0043] A multi-view social network community discovery device includes: at least one processor, at least one memory, and computer program instructions stored in the memory, which, when executed by the processor, implement the method as described.

[0044] Optionally, the memory is a computer-readable storage medium.

[0045] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0046] This invention provides a method, system, and device for discovering communities in multi-view social networks. It combines multi-view network data with clustering analysis algorithms to achieve user community identification. A low-pass filter is used to filter high-frequency noise to obtain a smooth representation, improving the accuracy and reliability of community identification in social networks. In the smooth representation fusion stage, an attention mechanism is used to fuse the smooth representations of each view to obtain a fused representation containing all view information. In the topology fusion stage, based on the fused representation and the fused Laplacian matrix, a trained MLP network based on the attention mechanism fusion method is used to determine the embedding representation. Ultimately, this achieves relatively stable and accurate social network user community identification, providing data support for content delivery, advertising, and other applications. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a schematic diagram of a multi-view social network community discovery method provided by the present invention;

[0049] Figure 2 This is a schematic diagram illustrating the principle of a multi-view social network community discovery method provided by the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] The purpose of this invention is to provide a method, system, and device for discovering communities in multi-view social networks, which can achieve stable and accurate identification of social network user groups.

[0052] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0053] like Figure 1 and Figure 2 As shown, the multi-view social network community discovery method provided by the present invention includes:

[0054] S101, acquire multi-view network data; the multi-view network data includes, but is not limited to: basic attribute information, social relationships, shared interests and hobbies, and interaction behavior records;

[0055] This also includes:

[0056] Preprocessing of multi-view network data includes filling in missing values ​​and removing outliers.

[0057] The preprocessed basic attribute information is represented as an X matrix.

[0058] Social relationships, shared interests, and interaction records are represented as an adjacency matrix A1, A2, ..., A1. M Both are used as input to the algorithm. Specifically, given data {X, A} of n users and m views. 1 A 2 ,...,A M}.in, d is the feature dimension of attribute X.

[0059] S102, use a low-pass filter to filter the multi-view network data to obtain a smooth representation H for each view. m Smoothness represents H m for:

[0060] Where I is the identity matrix, L m Let X be the Laplacian matrix of the m-th view, and let X be the matrix of basic attribute information. Let each represent the symmetric regularized Laplacian matrix of the m-th view, and k be the number of filtering layers. Let D be the normalized adjacency matrix of the m-th view. m Let be the degree matrix of the m-th view.

[0061] S103, utilize the attention mechanism to fuse the smooth representation H of each view. m The fusion is performed to obtain the fused representation.

[0062] S103 specifically includes:

[0063] The initial fusion representation is determined based on the initial weights.

[0064] Perform spectral clustering on the initial fusion representation to obtain the clustering result P, i.e., Query.

[0065] Performing spectral clustering on the smooth representation yields the clustering result Q. m Key m .

[0066] Based on clustering results P and Q m Update the initial weights to obtain the updated weights w. m The currently updated weight w m for:

[0067]

[0068] Where Value is set to 1, score(Q) m Q is calculated using the clustering result P as the true label and the clustering evaluation index ACC. m The clustering results score.

[0069] When the difference between the currently updated weight and the previously updated weight is less than the difference threshold, i.e. After the weights converge, use the currently updated weights w n Determine fusion representation Fusion representation for:

[0070]

[0071] Where n is the number of updates.

[0072] When the difference between the currently updated weight and the previously updated weight is not less than the difference threshold, return to the step of determining the initial fusion representation based on the initial weight, until the difference between the currently updated weight and the previously updated weight is less than the difference threshold.

[0073] S104. The Laplacian matrix corresponding to each view is fused using the learned weights to obtain the fused Laplacian matrix.

[0074] S104 specifically includes:

[0075] Using formula Determine the fusion Laplacian matrix.

[0076] in, Let t = {t1, t2, ..., t} be the fused Laplace matrix. M} is the learning matrix, t m The learning weights corresponding to the m-th view are given by w. n To initialize, L m Let be the Laplacian matrix corresponding to the m-th view, where M is the number of views in the multi-view network data.

[0077] S105, Based on the fused representation and the fused Laplacian matrix, the embedded representation is determined using a trained MLP network based on the attention mechanism fusion mechanism.

[0078] S105 specifically includes:

[0079] The fused representation and the fused Laplacian matrix are fused.

[0080] The fused data is then filtered to obtain the final representation H. α ,in,

[0081] According to the final representation H α A pre-trained MLP network based on an attention-based fusion mechanism is used to determine the embedding representation Z. Z = f(H) α ;W)=H α W; where W is the matrix of learnable weights of the MLP network, f(H α ;W) represents the mapping of the MLP network.

[0082] The training process of an MLP network based on an attention-based fusion mechanism is as follows:

[0083] Construct the training set and loss function based on the embedding representation Z; determine the similarity matrix based on the embedding representation Z. After obtaining the similarity matrix S, sort it in descending order according to the similarity score. ij It is a node pair (v i ,v j The ranking is based on the required positive and negative sample ratio η. pos ,η neg Let the positive sample with the lowest similarity be r. pos Let the negative sample with the highest similarity be r. neg Therefore, node pairs (v) are generated. i ,v j The tag for ) is:

[0084]

[0085] Update the parameters and learned weights of the MLP network using the Adam optimizer.

[0086] S106, clustering the embedding representation to obtain the final clustering result, realizing community discovery in social networks.

[0087] The resulting clustering divides users into several clusters, resulting in high similarity within clusters and low similarity between clusters. This ensures that users within the same category share a high degree of similarity in interests, hobbies, and desired items. In subsequent advertising and short video push notifications, this can be combined with information about other users within the same category to provide more targeted recommendations.

[0088] Corresponding to the above method, the present invention also provides a multi-view social network community discovery system, comprising:

[0089] The multi-view network data acquisition module is used to acquire multi-view network data, which includes: basic attribute information, social relationships, relationships based on shared interests, and records of interactive behaviors.

[0090] The smooth representation determination module is used to filter the multi-view network data using a low-pass filter to obtain a smooth representation of each view.

[0091] The fusion representation determination module is used to fuse the smooth representation of each view using an attention mechanism to obtain a fused representation.

[0092] The Laplacian matrix fusion module is used to fuse the Laplacian matrix corresponding to each view using learned weights to obtain a fused Laplacian matrix.

[0093] The embedding representation determination module is used to determine the embedding representation based on the fused representation and the fused Laplacian matrix, using a trained MLP network based on the attention mechanism fusion mechanism.

[0094] The community detection module is used to cluster the embedded representations to obtain the final clustering results, thereby realizing community detection in social networks.

[0095] In order to execute the methods corresponding to the above embodiments and achieve the corresponding functions and technical effects, the present invention also provides a multi-view social network community discovery device, including: at least one processor, at least one memory, and computer program instructions stored in the memory, which implement the method when the computer program instructions are executed by the processor.

[0096] The memory is a computer-readable storage medium.

[0097] Based on the above description, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned computer storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0098] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0099] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for discovering communities in a multi-view social network, characterized in that, include: Acquire multi-view network data; The multi-view network data includes: attribute information, social relationships, shared interests and hobbies, and interaction records; The multi-view network data is filtered using a low-pass filter to obtain a smooth representation of each view; The smooth representations of each view are fused using an attention-based fusion mechanism to obtain a fused representation; The Laplacian matrix corresponding to each view is fused using the learned weights to obtain the fused Laplacian matrix; Based on the fused representation and the fused Laplacian matrix, a trained MLP network based on the attention mechanism fusion mechanism is used to determine the embedding representation; Clustering the embedding representation yields the final clustering result, enabling community detection in social networks; The process of fusing the smooth representation of network data using an attention mechanism to obtain a fused representation specifically includes: The initial fusion representation is determined based on the initial weights; Perform spectral clustering on the initial fusion representation to obtain the clustering results. ; Spectral clustering is performed on the smooth representation to obtain the clustering results. ; Based on clustering results and clustering results Update the initial weights to obtain the current updated weights; When the difference between the currently updated weight and the previously updated weight is less than the difference threshold, the fused representation is determined using the currently updated weight; When the difference between the currently updated weight and the previously updated weight is not less than the difference threshold, return to the step of determining the initial fusion representation based on the initial weight, until the difference between the currently updated weight and the previously updated weight is less than the difference threshold. The process of fusing the Laplacian matrix corresponding to each view using learned weights to obtain the fused Laplacian matrix specifically includes: Using formula Determine the fusion Laplacian matrix; in, For the fused Laplace matrix, , is the learning matrix, For the first The learning weights corresponding to each view For the first The Laplacian matrix corresponding to each view. This represents the number of views in the multi-view network data. The step of determining the embedding representation based on the fused representation and the fused Laplacian matrix, using a trained MLP network based on an attention-based fusion mechanism, specifically includes: The fused representation and the fused Laplacian matrix are fused; The fused data is then filtered to obtain the final representation. ; According to the final statement A pre-trained MLP network based on an attention-based fusion mechanism is used to determine the embedding representation. .

2. The method for discovering communities in a multi-view social network according to claim 1, characterized in that, The process of acquiring multi-view network data further includes: Preprocessing of multi-view network data includes filling in missing values ​​and removing outliers. The preprocessed attribute information is represented as Matrix form; Social relationships, shared interests, and interaction records are represented as an adjacency matrix. .

3. The method for discovering communities in a multi-view social network according to claim 1, characterized in that, The training process of an MLP network based on an attention-based fusion mechanism is as follows: According to the embedding representation Construct the training set and loss function; Update the parameters and learned weights of the MLP network using the Adam optimizer.

4. A multi-view social network community discovery system, used to implement the multi-view social network community discovery method as described in any one of claims 1-3, characterized in that, include: Multi-view network data acquisition module, used to acquire multi-view network data; The multi-view network data includes: attribute information, social relationships, shared interests and hobbies, and interaction records; The smooth representation determination module is used to filter the multi-view network data using a low-pass filter to obtain a smooth representation of each view. The fusion representation determination module is used to fuse the smooth representations of each view using an attention mechanism to obtain a fusion representation; The Laplacian matrix fusion module is used to fuse the Laplacian matrix corresponding to each view using learned weights to obtain a fused Laplacian matrix. The embedding representation determination module is used to determine the embedding representation based on the fused representation and the fused Laplacian matrix, using a trained MLP network based on the attention mechanism fusion mechanism. The community detection module is used to cluster the embedded representations to obtain the final clustering results, thereby realizing community detection in social networks.

5. A multi-view social network community discovery device, characterized in that, include: The method comprises at least one processor, at least one memory, and computer program instructions stored in the memory, which, when executed by the processor, implement the multi-view social network community discovery method as described in any one of claims 1-3.

6. The multi-view social network community discovery device according to claim 5, characterized in that, The memory is a computer-readable storage medium.

Images