Peer Review #3 of "An improved two-stage label propagation algorithm based on LeaderRank (v0.3)"
2022
To solve the problems of poor stability and low modularity (Q) of community division results caused by the randomness of node selection and label update in the traditional label propagation algorithm, an improved two-stage label propagation algorithm based on LeaderRank was proposed in this study.In the first stage, the order of node updating is determined by the participation coefficient (PC).Then, a new similarity measure is defined to improve the label selection mechanism so as to solve the problem of label oscillation caused by multiple labels of the node with the most similarity to the node.Moreover, the influence of the nodes is comprehensively used to find the initial community structure.In the second stage, the rough communities obtained in the first stage are regarded as nodes, and their merging sequence is determined by the PC.Next, the nonweak community and the community with the largest number of connected edges are combined.Finally, the community structure is further optimized to improve the modularity so as to obtain the final partition result.Experiments were performed on 9 classic realistic networks and 19 artificial datasets with different scales, complexities, and densities.The modularity and normalized mutual information (NMI) were used as evaluation indexes for comparing the improved algorithm with dozens of relevant classic algorithms.The results showed that the proposed algorithm yields superior performance, and the results of community partitioning obtained using the improved algorithm were stable and more accurate than those obtained using other algorithms.In addition, the proposed algorithm always performs well in nine large-scale artificial data sets with 6000 to 50000 nodes and three large realistic network datasets, which verifies its computational performance and utility in community detection for large-scale networks.