Fast Search to Detect Communities by Truncated Inverse Page Rank in Social Networks

Abstract

Personalized PageRank is a useful technique for identifying a community with respect to a given node set. To obtain the overall community structure of the network, personalized PageRank should be executed amounts of times, which is prohibitive in massive networks. In this paper, to avoid useless and repeated computation, we propose a method that detects communities by truncated inverse PageRank. An efficient algorithm for computing the rank score in truncated inverse PageRank is devised. The computation only utilizes local information of the corresponding node. Rank score between local neighbors is regarded as a measure to select initial seed for each community. Inspired by work on seed set expansion, after excluding the nodes that are clearly true negative in seed set candidates, a seed set is initialized. Community expansion with rejudgement ensures that our method can detect community efficiently and precisely. Extensive experiments on different types of networks demonstrate the high performance of our method in terms of time and quality.