Research And Implementation Of Group Behavior Evolution Analysis Technology Integrating Semantic Network

Nowadays,people often communicate through social networks,and form groups based on social relationships and interests.Negative groups among them often lead to bad offline events.It is increasingly important to analyze and reveal the behavioral evolution rules of these groups to assist network emergency management.However,due to the sparse semantics and complex characteristics of negative social groups,the mainstream social group research methods such as academic networks and email communication networks are not suitable and have low analysis accuracy,making it difficult to support network emergency management decisions.To this end,combined with the data characteristics of negative groups in social networks and the semantic network,this paper proposes a group behavior evolution analysis technology that integrates the semantic network,as follows:First,find the negative user groups of social networks.Aiming at the problem that the semantics of social network texts are sparse,which leads to the low accuracy of group discovery,a group discovery method fused with semantic network is proposed.The text is semantically expanded by combining word importance rules,pre-trained word vectors and semantic networks,and a hierarchical attention mechanism is integrated on the basis of mainstream text recognition models.The background knowledge of the semantic network is integrated with the original text to detect negative texts.Combined with user social relations and group discovery algorithm,group discovery is performed on the set of users who have posted negative texts,and the accuracy of group discovery is improved.Second,analyze the behavioral evolution of groups from both microscopic and macroscopic perspectives.Aiming at the problem of low accuracy of intra-group interaction behavior prediction from a microscopic perspective,mainstream methods ignore user behavior evolution and construct a single feature.A prediction model of intra-group interaction behavior based on graph neural network and feature engineering is proposed to improve the prediction accuracy.In view of the problems of large noise and single dimension of group portraits from a macro perspective,based on user text and interaction data between users,combined with part of speech and word vectors and the topological characteristics of group behavior,the group portrait at each moment is depicted to represent the group state for tracking the evolution of group behavior.Third,design and implement the evolution analysis system of group behavior.It realizes group behavior analysis technology functions such as group discovery,intra-group interaction behavior prediction,group portrait characterization,and the visualization of group behavior rules.In order to verify the effectiveness of the group behavior evolution analysis method proposed in this paper,an experimental verification scheme is designed based on the Enron public dataset and the Twitter self-built dataset.The experimental results show that compared with the baseline model,the text recognition model in this paper has an increase of 0.47%in the F1 value,and the selected Louvain group discovery algorithm is 0.202 higher than the baseline algorithm in terms of modularity index;the intra-group interaction behavior prediction model in this paper is in Enron and Twitter datasets,the AUC indicators of the baseline model are 4.74%and 9.75%higher,respectively;and the content keywords extracted by the group portrait characterization method are improved by 3%and 2.2%in F1@10 and F1@20 respectively compared with the baseline method,and the group interaction rate is introduced to reveal the interaction frequency of group members.