Research On Dynamic Community Detection Method Of Incremental Node Embedded Node Embedding Influence-based

With the complication of interpersonal relationships,the scale of social networks has become larger and more complex,and the detection of community structures in complex networks has become increasingly important.With the application of deep learning technology,the research of community detection based on graph neural networks has also attracted more and more attention.On the one hand,the low quality of the graph structure corpus reduces the quality of the node vector representation,and the accuracy of the static community detection results based on the node decreases;on the other hand,the static community detection results change continuously over time.Therefore,this static community detection result is used as a network snapshot at the initial moment of the evolution process of the dynamic network structure.When calculating the community detection result at each moment,the original community detection method based on node representation is no longer applicable to dynamic community detection,and except for the original In the evolution process,there are other factors that will affect the final dynamic community results,so the problem we need to solve is to find an effective community detection algorithm for node representation in static community detection,and to detect in dynamic communities.At the same time,continue to use the new node representation to build a new model to improve the accuracy of the dynamic community detection results,so that it can restore the community structure of the real network.Aiming at the above problems,this paper proposes a community detection algorithm based on improved node representation and an incremental dynamic community detection model based on modularity,as well as related calculation methods and theoretical principles.The main work of the thesis includes:1)A community detection algorithm based on improved node representation is proposed.The community detection algorithm based on the improved node representation is divided into three phases: the random walk generation corpus phase,the improved node representation learning phase and the static community detection phase.In the stage of generating random corpora,a new search strategy was defined by combining the network topology information and the influence attribute information of nodes.The quality of the corpus constructed by the generated walking sequence was improved.In the improved node representation learning phase,a new objective function is defined.When the model converges,a semantically rich node vector representation can be obtained.In the static community detection phase,the many-to-many relationship between nodes and communities is considered comprehensively,that is,the same nodes in different sequences can belong to different communities,and different communities in the same sequence can contain different nodes.On the premise of the home community of the central node,the home community of its neighbors is predicted,and the community belonging status of all nodes can finally be obtained,and the effect of community detection is achieved.2)An incremental dynamic community detection model based on modularity is proposed.The model first determines the changes in the community caused by the changed nodes based on the improved definition of the modularity increment.It is based on the coarse-grained judgment of the topology structure.This step can realize the coarse-grained location of the changed nodes and the number of communities..Next,according to the new node and the removed node,the dynamic community detection algorithm of the newly added node and the dynamic community detection algorithm of the removed old node are respectively executed,and combined with the community changes that were initially located in the first step,the During the evolution,only the newly added node vector is used when the number of communities is constant,and the incrementally trained node vector and community vector are used when the number of communities is changed.When the evolution of removing old nodes occurs,incremental training node vectors are used when the number of communities is constant,and incremental training node vectors and community vectors are used when the number of communities is changed.In this way,through the second step,it is possible to achieve a fine-grained calculation of the node's community belonging probability,and then fine-grained adjustments to the node's community belonging situation.The error improves the accuracy of the final dynamic community detection and restores the community structure of the real network.3)In the simulation experiments,the algorithms and models proposed in this paper are compared and analyzed with several typical algorithms.It is verified that the community detection algorithm based on improved node representation proposed in this paper can improve the quality of the corpus,enrich the semantic information contained in the trained node vector,and improve the accuracy of community detection results.It also verifies the incremental dynamic community detection model based on modularity,which can be adjusted by coarse-grained partition positioning and fine-grained computing node belonging probability,which reduces the error generated by the community detection results at each moment and improves the final community.The accuracy of the detection results restores the community structure of the real network.