Multi-step locally expansion detection method using dispersed seeds for overlapping community

Abstract

The local expansion method is a novel and promising community detection algorithm. Just based on part of network information, it can detect overlapping communities effectively, but some problems exist such as seed node aggregation, poor quality and inaccurate community coverage. Therefore, we propose a local expansion overlapping community detection algorithm based on dispersed seeds. There are four essential parts of this algorithm: 1) We firstly generate non-overlapping partitions of the network, and locate seed nodes with the largest influence in their own partition by using a new index of node influence, which combines the information centrality of nodes and the number of k-order neighbors. 2) Secondly, on the condition of the neighborhood overlap measure maximization, seed nodes merge unseeded nodes to generate a preliminary seed community; 3) Then based on the community conductance gain, the allocated nodes are screened and the free nodes are assigned to the seed community; 4) In the end, a node-community similarity based on common connection edge is proposed to re-allocate new free nodes and obtain the final community structure. This method can make the community distribution more proper and the coverage more reasonable. The experimental results on some artificial data and real network data show that the algorithm performs well on overlapping community indicators such as EQ and ONMI, while the community detection results are more stable.