Abnormal user detection based on instant messages

Abstract

Instant messaging (IM) tools have been widely used in peoples' daily life. We study how to detect the identities of IM users from their chatting text. The abnormal detection model is employed to detect the identities of IM users. We use the topic model to find the relations between function words of chatting text, and extract the topic features to represent chatting text To improve the accuracy, we combine topic features with word based features to train the detection model, and achieve good experimental results.