Behavior deviation: An anomaly detection view of rumor preemption

Abstract

Since information can spread rapidly and widely more than ever on Online Social Networks (OSNs), they have become new hot beds of false rumor diffusion. Due to the potential harm these false information may bring to the public, false rumor detection has become a significant but challenging research topic. While previous research work mostly views it as a classification task, we treat it as an anomaly detection problem. In this paper, false rumors are viewed as anomalies and we perform Factor analysis of mixed data (FAMD) on our proposed features to detect these anomalies. Two strategies based on Euclidean distance and Cosine similarity are proposed to describe the deviation degree. A rank based on deviation degree is computed which can facilitate further rumor detection. We show our method can achieve good performance and can shed light on automatic detection of false rumors on OSNs.