Identifying user behavior on Twitter based on multi-scale entropy

Abstract

Twitter as an online social network is used for many reasons, including information dissemination, marketing, political organizing, spamming, promotion, conversations and so on. Characterizing these activities and categorizing users is a challenging task. Traditional user classification models are based on individual user's profile information such as age, location, register time, interests and tweets, which have not considered the whole complexity of posting behavior. In this paper we introduce Multi-scale Entropy for analyzing and identifying user behavior on Twitter, and separate users to different categories. We have identified five distinct categories of tweeting activity on Twitter: individual activity, newsworthy information dissemination activity, advertising and promotion activity, automatic/robotic activity and other activities. Through the experiment we achieved good separation of different activities of these five categories based on Multi-scale Entropy of users' posting time series. The method based on Multi-scale Entropy is computationally efficient; it has many applications, including automatic spam-detection, trend identification, trust management, user-modeling in online social media.