Evaluation repeated random walks in community detection of social networks

Abstract

The repeated random walks algorithm (RRW) is a graph clustering algorithm proposed recently. RRW has been shown to achieve better performance on functional module discovery in protein-protein interaction networks than Markov Clustering Algorithm (MCL). There is however little work applying RRW to community detection in social networks. We ran RRW on some real-world social networks that are well-documented in the literature. We then analyzed the impact of different parameters on the quality of clustering, by using a set of cluster metrics. We also compared RRW with two other random walk based graph clustering algorithms. Our experiments showed that the RRW algorithm achieved higher precision but lower modularity. The experiments also revealed some weaknesses of the RRW algorithm, such as higher running cost, and “discarding nodes” method in its post-process stage, which greatly affects the quality of clustering.