Research On Target Group Detection In Social Network

With the development of the mobile Internet,social network applications have rapidly become popular,covering most people.Social networks facilitate people's lives,but some harmful groups have also appeared in social networks.These harmful groups use social network applications to communicate and communicate,which is likely to have serious adverse consequences.Group discovery is one of the important topics in social network research.Research group findings can help understand the structural characteristics of groups in social networks,the mechanism of group formation,and can guide and control harmful groups based on group findings to reduce the harmful effects of these group behaviors.This article focuses on how to quickly and accurately discover all specific group nodes in social networks based on the initial nodes of the group,and provides a new idea for the discovery of specific groups.A social network graph is a typical graph data structure,and its nodes have a large amount of text content,which represents the attribute information of the node.This paper proposes a specific group discovery algorithm based on graph neural network,which can aggregate graph structure information and graph node attribute information well.This study uses word frequency analysis to model users using social network text,conducts specific group attribute subspace mining on initial nodes,uses focus attributes on attribute subspaces to perform specific group discovery,and uses positive examples and unlabeled sample learning(Positive-unlabeled learning)bagging algorithm(bagging),a series of graph convolutional network classifiers are constructed,and the average result of these classifiers is taken as the group discovery result.Aiming at the situation that the number of initial nodes is too small,a specific group discovery algorithm based on pseudo-tag strategy is proposed.Use the graph autoencoder based on the graph attention mechanism to embed social network nodes into a lowdimensional vector space,run a clustering algorithm on the node embedding vector,select highly reproducible nodes to assign pseudo labels,and use pseudo label nodes as sample training graphs Convolutional network classifiers perform specific group discovery.Experiments were carried out on different types of networks,and it ws found that the two algorithms can achieve an accuracy rate exceeding 0.9 on a simple network,and can achieve better results in a complex network.