A framework for spotting anomaly

Detecting anomaly nodes from graphs is an important objective in many applications ranging from social networks to World Wide Web. Recently several methods have been proposed to address this problem. A limitation of most of these methods is that they are based on the random walk of the graph, and often fail to be effective. In this paper, we propose a new framework to detect anomaly nodes within a graph. The approach relies on a novel mapping strategy that maps nodes to a multidimensional space wherein sparse areas in the mapped space correspond to anomaly nodes. The algorithm then spots the sparse regions in the mapped space to output the anomaly nodes of the graph. Our experiments on real datasets prove that the proposed framework is effective.