A Community Detection Algorithm Based on Jaccard Similarity Label Propagation

Due to the randomness of label propagation algorithm, the stability is poor and the accuracy is low for community detection results in complex networks. In order to solve the problem, this paper proposes a novel community detection algorithm based on Jaccard similarity label propagation. Firstly, the Jaccard similarity is used to measure nodes importance. Then, the importance of nodes is utilized to reduce the randomness in label selection. Finally, nodes with the highest importance are selected to update labels in iteration, which improves the stability for community detection. Stability and accuracy of data sets of real networks are measured by modularity and normalized mutual information, respectively. Experimental results show that, the proposed algorithm for community detection results is more stable and more accurate than LPA, LPA_SI and KLPA algorithms in the cases of near linear time complexity.