Research On Overlapping Community Detection Methods For Complex Networks Based On Node Influence Expansion

Nowadays,it is more common than ever to analyze the huge amount of data generated in social media,and to use it effectively to facilitate people’s lives,such as recommendation engines,marketing and crime detection.Although there is no uniform definition of the concept of "community",it has been observed and studied over a long period of time that nodes within a community are tightly connected,and nodes within a community are sparsely connected to nodes outside the community.Communities are also an important topological feature of complex networks,which can support the study of structural features of networks and guide real-world applications.Community detection can precisely discover relationships between data entities,probe network topology,and provide theoretical support for interest prediction,friend recommendation,and event evolution.The seed expansion algorithm has the advantages of robustness and efficiency.It is an important community discovery method.The main process of seed expansion algorithm is divided into two stages,firstly the first stage is to find the seed as the initial expansion node,and finally the second stage is to select the appropriate method to expand the seed node based on the already selected seed node to get the appropriate community structure.This study finds that existing community discovery methods based on seed expansion mostly ignore the influence and role of nodes’ own influence in community formation when seeds are selected,which will lead to the problems of low community structure cohesion and weak structural representation.In this paper,we propose a new community discovery method based on the combination of node influence and node global structure information for seed selection.The core idea of the method is to first filter the more influential nodes,which are the more important nodes in the community,and to eliminate the bridging nodes(nodes located at the edge of the community to play a bridging role)of the seed nodes.In the community generation phase,the idea of sequential expansion is used,which is more likely to reduce the expansion of low-influence nodes to high-influence nodes,which makes the algorithm more consistent with the actual influence propagation model.This paper accomplishes the following:Constructing the network node-to-node propagation probabilities.Before starting the experiments,this study applies the Single-cascade model(SCM)model to the experimental data on the network for planning analysis to provide prerequisites for calculating the influence of the nodes,which further supports the seed selection in the seed expansion algorithm.Constructing the seed mass functionand and filter seed nodes.In the seed expansion algorithm,it is very important to select high quality seed nodes,and in this study a combination of node influence and node eccentricity is used to screen seeds in the seed selection stage,and node influence strength is proposed.Constructing seed expansion community functions and starting community expansion.In the community discovery algorithm of seed expansion,the process of seed node expansion into a community is also an extremely important stage.In this paper,we combine the node influence with the network structure to expand the community,and propose the influence-weighted local expansion factor.Experimental results and analysis.The first part introduces the basic information of the experiment,including the experimental data set and the experimental environment;the second part of the experiment verifies the performance of the NIEM algorithm in terms of the number of seeds in the seed selection phase and the influence of nodes in the expansion phase;the third and fourth parts of the experiment compare the performance differences between the NIEM algorithm and various classical community recognition algorithms on synthetic and real networks;finally,compare Finally,the experimental metrics of NIEM algorithm and various classical community recognition algorithms on DBLP network are compared.The experimental results verify the effectiveness and feasibility of this research algorithm.