An adaptive random walk sampling method on dynamic community detection

Abstract

With the change of lifestyle and interests, the people’s social activities have a dynamic changing tendency. Therefore, the static community could not reflect the real activities. For the ’community’ in the social network is the aggregate of people’s activities, thus the dynamic community could be detected by simulating the individual freewill. The individual tends to get in touch with the closest friends. By that a direction from one node to its closest nodes can be obtained, and the formed directed network could easily find out the communities. It is different from the traditional community detection policies, which only consider the global topological structure of the social network. Accord to the theory above, we designed the RWS (Random Walk Sampling) method to detect the overlapping communities, utilizing the random walk method to find the closest friends for each node. As the topological structure changing, the proposed ARWS (Adaptive Random Walk Sampling) could make the impacted nodes find out the new closest friends and the changed communities adaptively. The ARWS only update the impacted nodes and communities as the dynamic events occurring, while the traditional dynamic community detection methods need to break up and restructure the communities after the topology changing, because the tradition methods are based on the global topological structure. Therefore, the ARWS has a lower cost than the traditional methods. Furthermore, the ARWS focus on the individual, fitting to the decentralized computing framework, such as distributed computation and cloud computing. That is the trend of the artificial intelligence.