Threshold Random Walkers for Community Structure Detection in Complex Networks

There exist large amounts of complex networks in different areas nowadays, which have aroused great interest in detecting community structures. Although diverse community detection algorithms have been proposed, most of them perform poorly in large scale complex networks. According some social principles, we proposed a scalable Community Detection method based on Threshold Random walkers, which is called CD-TRandwalk. CD-TRandwalk selects active nodes with high degree as seed nodes, and detects the core communities through random walkers according to predefined thresholds at first. Because the threshold random walkers start from the active seed nodes and only randomly walk to those nodes which association degrees are larger than a given threshold, the processes of detecting core communities work quickly. After that, the remaining non-core nodes are allocated into the core communities according their common degrees between these nodes and the core communities with a voting strategy. Compared with some other community detection algorithms such as Affinity Propagation (AP), Walktrap, Newman Fast, and ComTector in several social networks, the experimental results show that CD-TRandwalk is faster than the other methods without worse quality of community detection quality. Furthermore, CD-Trandwalk is adaptable to large scale networks and unbalance networks. CD-TRandwalk also has some other advantages, such as it is unsupervised and not need to set the community number beforehand, and it only needs local information of the networks to support local community detection.