Weighting links based on edge centrality for community detection

Link weights have the equally important position as links in complex networks, and they are closely associated with each other for the emergence of communities. How to assign link weights to make a clear distinction between internal links of communities and external links connecting communities is of vital importance for community detection. Edge centralities provide a powerful approach for distinguishing internal links from external ones. Here, we first use edge centralities such as betweenness, information centrality and edge clustering coefficient to weight links of networks respectively to transform unweighted networks into weighted ones, and then a weighted function that both considers links and link weights is adopted on the weighted networks for community detection. We evaluate the performance of our approach on random networks as well as real-world networks. Better results are achieved on weighted networks with stronger weights of internal links of communities, and the results on unweighted networks outperform that of weighted networks with weaker weights of internal links of communities. The availability of our findings is also well-supported by the study of Granovetter that the weak links maintain the global integrity of the network while the strong links maintain the communities. Especially in the Karate club network, all the nodes are correctly classified when we weight links by edge betweenness. The results also give us a more comprehensive understanding on the correlation between links and link weights for community detection.