An Overlapping Community Detection Algorithm Based On Complete Subgraph And Label Propagation

Overlapping community discovery can help people analyze and understand complex networks,and has become a hot topic in data mining research.As new overlapping community discovery algorithms are constantly proposed,the real-world application situation will continuously increase the time performance and stability requirements of the algorithm.COPRA algorithm is a classic overlapping community detection algorithm based on label propagation.The algorithm uses label propagation to judge node ownership,and introduces the concept of attribution coefficients and thresholds,which can effectively complete the detection of overlapping communities.The COPRA algorithm has good temporal performance,but it is still weak when dealing with large-scale networks.In addition,the stability of the COPRA algorithm is poor,and the results of community discovery have great randomness.This paper proposes a complete subgraph and label propagation overload algorithm(CLPOA)based on complete subgraph and label propagation.The algorithm is mainly based on the following two key points:(1)The COPRA algorithm assigns each node an independent label in the initial label allocation stage.This allocation method directly leads to a large time overhead of the COPRA algorithm.This paper analyzes the relationship between community structure and complete subgraphs.In the CLPOA algorithm label initialization process uses complete subgraphs to replace community models.Through searching all complete subgraphs in the network to get the initial community structure and assign unique labels to each community.The program theoretically has better time performance.(2)The COPRA algorithm introduces a random selection strategy in the label discarding stage of the label propagation stage.The algorithm will choose one of the multiple labels that meet the conditions to retain the probability.This strategy directly leads to the instability of the algorithm.The CLPOA algorithm proposes the contact frequency as the reference value of the tag's influence on the node.By comparing the contact frequency to the tag selection,the execution conditions of the random selection strategy are more stringent,the randomness of the algorithm is also reduced,and the experimental result is also more stable.In this paper,two standard network data sets of the Dolphin Community Network and the American College Football League were selected to conduct the experimental test.The results show that the CLPOA algorithm has better algorithm stability and time while maintaining the same quality of community classification as the COPRA algorithm.performance.