Dynamic Community Detection Algorithm Based On Graph Embedding And Nonnegative Matrix Factorization

Complex networks can model complex systems in the real world.Community detection is one of the hotspots of complex network analysis.It aims to detect a community structure where the nodes within the same community are closely connected and the nodes between different communities are sparsely connected.Community detection provides an opportunity to study complex networks’ potential operation mechanisms and the interaction between modules.However,most community detection methods are based on static networks.The real world is changing all the time.Static network community detection cannot track and represent the community evolution process of dynamic networks.Therefore,this paper proposes a dynamic community detection algorithm based on third-order temporal local smoothing and a dynamic community detection algorithm based on edge dynamic smoothing.Finally,the dynamic community detection algorithm is applied to the pathogenic module detection of breast cancer dynamic networks.The main work of this paper can be summarized as follows:(1)From the perspective of "community-level smoothing," the community smoothing of dynamic network needs to judge whether the members in the same community are the same at successive time steps.In this paper,a new third-order temporal local smoothing framework based on graph embedding is proposed.Introducing a graph embedding matrix instead of the original adjacency matrix can reduce the algorithm’s time complexity and preserve the high-order topology information of the original network.Introducing the third-order temporal local smoothing can better describe the evolution process of the dynamic network and make the clustering result of the dynamic network more robust.The experimental results show that the model can effectively track and detect the network evolution process.(2)From the perspective of "edge smoothing," the edge smoothing of dynamic network needs to enhance the clustering information brought by the common edge of three adjacent snapshots and eliminate the noise edge’s clustering interference.This paper proposes a new dynamic community detection framework based on graph embedding and edge smoothing.The joint learning framework makes the process of graph embedding learn better features for clustering.Simultaneously,graph clustering tends to get better results under the guidance of the features of graph embedding.Edge clustering can effectively analyze the dynamic community’s evolution process from a more subtle angle of edge dynamics.Experimental results show that the model can get a good embedded representation of the graph and get accurate and robust community detection results.(3)In dynamic networks,besides topology information,there is also the graph’s attribute information.In this paper,a framework of joint decomposition of topological information and attribute information of dynamic graph is proposed,applied to the detection of cancer dynamic module.Through joint deterioration,the topological information and attribute information of the dynamic graph is fully fused.Through a large number of experiments on real breast cancer networks,our algorithm can accurately capture the pathogenic gene modules of some known breast cancer and provide a lot of reference information for the pathogenesis of breast cancer.The dynamic community detection algorithm proposed in this paper is an effective exploration of dynamic network community detection.The proposed algorithms provide many new ideas for the field of dynamic complex network analysis and have certain theoretical significance and application value.