Research On Event Detection Algorithm Based On High-order GAT In Social Heterogeneous Information Network

Social networks have become an indispensable part of production and life.Event detection is of great significance to public opinion monitoring,hotspot tracking,online confrontation,and national information security.Social network text fragmentation is severe,sparseness is high,and event elements contain heterogeneity,the existing methods cannot properly represent events.The rise of HIN provides new ideas and methods for existing problems,and can define new HIN for specific industries or fields.This article defines the SHIN for event detection.The SHIN＿Framework,an event detection architecture based on SHIN,is designed.On this basis,an event detection algorithm Sim GAT based on higher-order Graph Attention Network in a social heterogeneous information network is proposed.Specifically,the work of this article is as follows.First,the SHIN＿Framework event detection architecture based on SHIN is designed.For the heterogeneity of event elements,extract event elements from the original data information,construct SHIN to represent social events,select Meta-Path from them and extract node semantic features for clustering.Then,according to the above framework,SHIN is constructed,and a high-level GAT event detection algorithm Sim GAT is proposed in the social heterogeneous information network.Using the similarity calculation method Path Sim to calculate the similarity between nodes to construct an adjacency matrix,measure the weight of different nodes,and then improve GAT.Use the Attention mechanism of the graph attention layer to further explore the content-based node interaction and increase the stability of the feature on the basis of多头注意力.Secondly,for the effective feature representation,the high-order GAT model Sim GAT is used to further extract the information in the high-order neighborhood of the node,and gradually enhance the characteristics of the initial node.The multi-layer improved GAT model is combined back and forth to collect deeper embedding of nodes and neighbor nodes,and more fully and effectively extract the node information and semantic information in SHIN.Finally,the features extracted by Sim GAT algorithm are used as the input of density-based DBSCAN clustering algorithm to obtain clustering clusters.Each cluster represents a detected social event.Social event detection experiments on real data sets verify the effectiveness of the proposed algorithm.