CN-117194808-B - Social network propagation backbone structure discovery method based on association analysis

CN117194808BCN 117194808 BCN117194808 BCN 117194808BCN-117194808-B

Abstract

The invention realizes a social network propagation backbone structure discovery method based on association analysis. The method comprises the steps of firstly inputting a social media specific topic data social network library, extracting user characteristics of a social network, then selecting key users and key user groups formed by surrounding neighbor users in a top-k social network by the user importance measurement module, inputting the key user groups into the transmission backbone structure learning module, and outputting the transmission backbone structure by the transmission backbone structure learning module, wherein the transmission backbone structure learning module comprises a user association relation sub-module and a user interactivity sub-module. Therefore, the method and the system can acquire the associated information of surrounding user groups while mining key users in the social network, and acquire the effect of information transmission backbone networks under specific topics of the social network.

Inventors

SUN QINGBIN
QIN JIAWEN
AN HUA
JI CHENG
Yang beining
Yuan Haonan
LI JIANXIN

Assignees

北京航空航天大学

Dates

Publication Date: 20260508
Application Date: 20230824

Claims (3)

1. A social network propagation backbone structure discovery method based on association analysis is characterized by comprising a user importance measurement module and a propagation backbone structure learning module, wherein a social media specific topic data social network library is firstly input, if no topic exists, the social media specific topic data social network library is directly ended, if the topic does not exist, an information propagation network corresponding to the topic is taken out, a social network is constructed, and further user characterization of the social network is extracted; The user importance measurement module inputs the user characterization into a user local importance sub-module and a user global importance sub-module respectively, and obtains the importance score of each user according to the user local importance and the user global importance; Then selecting a key user group formed by key users and surrounding neighbor users in the top-k social network, and inputting the key user group into the propagation trunk structure learning module; According to key users and neighboring users around the key users in the top-k social network selected in the user importance measurement module, the probability of the existence of connection between each pair of users is obtained through the user association relation sub-module and the user interaction sub-module, the association relation between the users is learned again, and a new adjacency matrix is obtained, so that a transmission backbone structure is formed; The method for extracting the information propagation network corresponding to the topic comprises the steps of firstly obtaining topic data of a preset time period based on topic keywords, obtaining related content by using a crawler technology through the preset time period and the keywords, and if the content is a forwarding and comment relation, obtaining a connecting edge between two corresponding user nodes, and taking social properties and self properties of the user as initial user characterization of the user nodes; The user importance measurement module is used for inputting a feature matrix X of user nodes in the social network and a relation matrix A among the user nodes, obtaining low-dimensional representation output of each node based on a graph convolution neural network according to a constructed social network structure and the user node features acquired from a social media platform, calculating the relevance of the node and the neighbor node and the information relevance of a user group in the whole social network, and outputting importance scores of each node; the implementation mode of the user local importance calculation submodule is that a graph is given Definition of The adjacency matrix of the graph is represented, Characteristic matrix of representation graph, design a discriminator Analyzing each node and its corresponding Correlation between order neighbors: Wherein, the Representing a set of all nodes in the graph, Representing nodes A kind of electronic device The set of all neighbor nodes of the rank, Representation of Medium-NOT node Is provided with a plurality of nodes, wherein each node is connected with a plurality of nodes, Representing nodes Is used for the feature vector of (a), Representing nodes A kind of electronic device The sum of the features of all the neighboring nodes of the rank, Is a discriminator, the MLP is a multi-layer perceptron, Representing activation functions, maximizing implementation during training ; Designing nodes by calculating the ability of each node to describe neighbor nodes, namely the discriminability of the nodes to local information Is a local importance score of: ; The user global importance computing sub-module is realized by the following steps of The order neighbor region is regarded as a sub-graph in the large graph, the center of the sub-graph is the node itself, and a discriminator is designed Computing nodes The association degree of the order neighbor information and the whole graph is specifically defined as: Wherein, the Representing a set of all nodes in the graph, Representing nodes A kind of electronic device The set of all neighbor nodes of the rank, Representation of Medium-NOT node Is provided with a plurality of nodes, wherein each node is connected with a plurality of nodes, Representing nodes A kind of electronic device The sum of the features of all the neighboring nodes of the rank, Feature vectors representing social networks Is a discriminator, the MLP is a multi-layer perceptron, Representing activation functions, the need to maximize during training The ability of each sub-graph to describe the global information, namely the discriminant of the sub-graph to the global information, is calculated, and the node is designed Node global importance score for center: 。
2. The method for discovering social network propagation trunk structure based on association analysis of claim 1, wherein the specific generation method of the key user group is that top-k key users are selected in the last layer of network according to the user importance measurement score, a trunk structure learning layer is designed, a new structure is relearned according to the existing nodes, and user nodes and low-dimensional characteristic representations thereof output by the user importance measurement module are input.
3. The social network propagation backbone structure discovery method based on association analysis of claim 2, wherein the propagation backbone structure learning module extracts key participation characters in key users, and inputs the key participation characters into the user association relation learning sub-module and the user interactivity analysis sub-module; The user association relation learning submodule is specifically implemented by learning similarity between two nodes by using a single-layer neural network, namely, connectable scores, which are defined as follows: Wherein Is a weight vector that can be learned and, Representing an activation function; The specific implementation mode of the user interactivity analysis submodule comprises the steps of defining a user interactivity matrix Wherein the method comprises the steps of , Representing the shortest distance of two nodes in the social network, Representing the euclidean distance between two nodes in potential space, by the following calculation: Wherein Representing nodes A low-dimensional vector representation of the feature, And Are super parameters; To make similarity scores comparable across different nodes, regularization is performed using a softmax function: Wherein, the , For a super parameter, a new sub-graph structure is obtained And updating the trunk structure adjacent matrix to use a new adjacent matrix when the neighbor information aggregation is carried out in the next picture volume layer, and outputting the last layer of the model to be the trunk structure of the social network propagation.

Description

Social network propagation backbone structure discovery method based on association analysis Technical Field The invention relates to the technical field of data mining, in particular to a social network propagation backbone structure discovery method based on association analysis. Background In recent years, with the development and popularization of internet technology, more and more netizens communicate by using social networks, so as to gradually change the ecology of network public opinion and promote the transfer of public speaking rights. In a social network, information is propagated along a social network formed by association relations among users through propagation behaviors among users, and each individual and different groups play a great role in network structure and function. Key users are special individuals that can affect to a greater extent the network structure and functionality formed during the information dissemination process. For example, micro-doctor V may accelerate the diffusion of facts or rumors in a social network. Therefore, in order to better guide the network public opinion, key users and related user groups in the social network need to be accurately discovered from a large number of users, and the development trend of the network public opinion is captured and predicted by utilizing a transmission backbone structure formed by the users and the association relations among the users. The mining of the propagating backbone structure includes an important research direction, namely mining opinion leaders in the network, which play an important role in mediating or filtering in the formation of mass-spreading effects. Many social network opinion leader recognition algorithms have been proposed today, classical approaches based on network topology include centrality, betweenness centrality, proximity centrality, feature vector centrality, etc. The most intuitive and most basic method is centrality, is simple and effective in calculation, but only reflects local information of a complex network, and does not consider own characteristics of nodes and interaction among the nodes. And each evaluation method is based on statistics and has specific limitations, so that the evaluation method cannot be well applied to all types of networks. There are also cluster analysis, pageRank algorithm, etc. based methods for identifying opinion leader nodes in a network, which are advantageous in that the model is simple and can be converged in a relatively short time when there are more samples. However, the scale of the clusters cannot be controlled, and the nodes possibly do not all have similar characteristics under the same category, and the PageRank algorithm assumes that the nodes have the same jump probability and has the defect of non-unique sequencing results. Some neural network models are based on a method for mining a key user set, and the problem that a trunk structure formed by key users and neighbors thereof and group information of key users in different levels cannot be fully mined exists. Aiming at the problems that the prior research cannot fully utilize social network information to cause inaccurate identification of key users and insufficient discovery of user association to influence the discovery of a social network propagation backbone structure, the invention provides a social network based on user interaction under the construction of a social media specific topic, wherein the characteristics of nodes are enriched through the inherent attributes and the social attributes of users in the social network, the importance of one user in the social network is comprehensively measured from the local importance and the global importance respectively, top-k key users are obtained, and the reconstruction interaction relation between the users is obtained through the characteristics of the users and the topological distance characteristics, so that the discovery task of the social network propagation backbone structure is realized. Disclosure of Invention The invention firstly provides a social network propagation backbone structure discovery method based on association analysis, which consists of a user importance measurement module and a propagation backbone structure learning module, wherein a social media specific topic data social network library is firstly input, if no topic exists, the social media specific topic data social network library is directly ended, if the topic does not exist, an information propagation network corresponding to the topic is taken out, a social network is constructed, and further user characterization of the social network is extracted. The user importance measurement module inputs the user characterization into a user local importance sub-module and a user global importance sub-module respectively, and obtains the importance score of each user according to the user local importance and the user global importance; Then selecting a key user group