Research On Methods Of Dynamic Social Network Community Detection Based On Graph Embedding

As one of the main research directions of social network analysis,dynamic community detection has become the focus area of researchers and is widely used in many fields such as public security,public health,recommender systems,network analysis,link prediction,and public opinion monitoring.At present,dynamic community detection based on graph embedding mainly faces two problems.On the one hand,the traditional incremental dynamic community detection only focuses on the local changes of nodes and edges,which ignores the impact of historical structure information on the current snapshot network,making the dynamic community detection not accurate enough.On the other hand,in the process of dynamic network representation learning,shallow-level graph embedding algorithms are often unable to mine deep-level node dynamic interaction information,and cannot well represent the spatiotemporal features of dynamic networks.In response to the above problems,this thesis takes dynamic social networks as the research object,takes accurate dynamic community detection as the research goal,and combines graph embedding technology to propose dynamic community detection methods based on graph embedding.The main work of this thesis includes the following aspects.(1)Incremental dynamic community detection based on temporal walk embedding.Aiming at the problem that the traditional incremental dynamic community detection cannot consider the influence of the historical structure on the current network,an incremental dynamic community detection based on temporal walk embedding is proposed.The method is divided into two stages.In the data processing stage,firstly,a snapshot network fused with historical edge weights is constructed by using the historical time network,and a temporal walk strategy is designed to extract the node sequence.Then,word embedding is used to learn the node embeddings.In the community detection stage,at first,the definition of incremental node is proposed,and the embedded information is used as the attributes of dynamic nodes to construct a dynamic attribute network.What’s more,the dynamic network is divided into communities using a community detection method with improved modularity.Finally,the method enables accurate and efficient community detection for both synthetic and real-world dynamic networks.(2)Community detection based on spatiotemporal graph embedding in dynamic networks.Since the shallow graph embedding technology cannot fully mine the dynamic interaction information between nodes,a method of community detection based on spatiotemporal graph embedding in dynamic networks is proposed from the perspective of the spatiotemporal evolution of the dynamic network.Above all,the graph convolution neural network is used to aggregate the spatial interaction characteristics of dynamic network nodes,and the recurrent neural network is used to share the weight parameters in the convolutional network.Then,the gated recurrent unit networks are introduced to learn the temporal characteristic information of nodes,so as to construct the spatiotemporal graph embedding information of the dynamic network.Finally,nodes are partitioned according to the embedding features of each temporal snapshot network using self-organizing maps algorithm.Experiments show that this method can effectively learn the spatiotemporal information of dynamic networks and improve the accuracy of community detection.(3)Community detection systems in social networks.Based on the object-oriented design idea,the community detection system in social network is implemented by using the modular design idea,combining with the existing research and the research results of this thesis.The system integrates a lot of static and dynamic social network datasets and community detection algorithms,and can carry out experimental validation and result analysis of the research results of this thesis,which provides experimental basis for further large-scale application and expansion.