Abstract:

Community detection is an important research topic of complex network analysis.The detection results help to understand the community structure of complex networks and provide support for downstream tasks, such as content recommendation, link detection.Considering the challenge of community detection in complex networks, a two-stage community detection algorithm based on label propagation (TS-LPA) was proposed.The TS-LPA algorithm quantified the propagation capability of nodes based on the extended neighborhood.Then a new evaluation index was proposed to measure the probability of influence between nodes using the information of nodes and the weight of edges in the network.Based on the calculation of node centrality, the algorithm determined the updating sequence of node labels and the selection strategy of seed nodes, which eliminated the instability of the algorithm in the updating process.The TS-LPA algorithm used the breadth-first propagation idea and introduced the second-stage label propagation method to improve the quality of community detection.When label began to spread, all neighboring nodes had an influence on the label of the related node.Meanwhile, the influence of neighboring seed nodes was added to complete label updating, in order to reduce the dominance of neighboring nodes on the updated node.The experimental results of different real data sets and synthetic data sets show that TS-LPA algorithm can eliminate randomness and show strong stability while effectively improving the quality of community detection.

Key words: community detection, label propagation, node centrality, breadth-first search