Community Detection Algorithm Based On Node Influence And Similarity

Systems in society and nature can be regarded as complex networks,such as interpersonal relationship network,football network,dolphin network,and so on.In complex networks,nodes represent individuals in the system,and edges represent the connections between individuals.In particular,as an important topological property of complex networks,community structure has attracted widespread attention from scholars at home and abroad in recent years.A large number of community detection algorithms have been proposed to analyze network topology.Based on how many communities a node could belong to,community detection algorithms can be divided into two categories: non-overlapping community detection methods and overlapping community detection methods.Among them,label propagation algorithms are the most commonly used community detection algorithms.In terms of non-overlapping communities,label propagation algorithms have the outstanding nature of simple operation and low time complexity,so they can be applied to large complex networks.However,they also have the disadvantages of poor robustness,monster communities and excessive divisions.In terms of overlapping communities,multiple labels may be assigned to one node to find overlapping community structure,but it still has the disadvantage of poor stability.To solve the problems above,this paper proposes a label propagation algorithm based on node influence and similarity.First,the seed nodes are selected and they are expanded into seed areas;then the label propagation is performed;finally,the non-overlapping structure is obtained by merging the communities.Meanwhile,a label propagation algorithm for overlapping communities is proposed.Based on the non-overlapping structure,this algorithm obtains all edge nodes in the network and finds key overlapping nodes from the edge nodes;Then,searching for overlapping nodes in the key overlapping node set to get the final overlapping community partition result by combining the original node labels.In the process of community detection above,network topology and real node attributes are combined to find more realistic results.Experiments on synthetic networks and real networks show that the algorithmsproposed in this paper to detect non-overlapping community structure and overlapping community structure are superior to the existing label propagation algorithms in terms of network partition quality.Particularly,the algorithms proposed in this paper have also achieved good results in large real networks,Tencent Weibo Network.And the formed community structure is more in line with the situation of real world and has important practical significance.