Edge Weight Method for Community Detection in Scale-Free Networks

In this paper, we proposed edge-weight based method to perform a community detection in scale-free networks. Our main idea is that communities are formed from edges. We believe that in scale-free networks, each edge is not equal. Edges that connect to hub nodes are more important and should be weighted more than edges that connect to fewer nodes. Therefore, our proposed method begins with edge weight calculation based on centrality value of each edge. Then, two steps community detection algorithm performs iteratively until the community detection result is balanced. The first step of our algorithm is aimed to extract communities that node degree follows power law (scale-free). While the second step is aimed to extract communities that node degree follows normal degree distribution. The benefit is that our method can work on both scale-free and non scale-free networks. Moreover, in scale-free networks, not all communities contain hub nodes nor node degree distribution follows the power law. For scale-free approaches, this can cause errors in community detection. On the other hand, our method can handle this case correctly because our algorithm contains both scale-free and non scale-free approaches. To evaluate our method, we use NMI - Normalized Mutual Information - to measure our results on both synthetic and real-world datasets comparing with both scale-free and non scale-free community detection methods. The results show that, our method can perform almost equal or better on both scale-free and non scale-free networks.