Community Based Spammer Detection in Social Networks

Abstract

Social networks (SNS in short) such as Sina Weibo have now become one of the most popular Internet applications. A major challenge to social networks is the huge amount of spammers, or fake accounts generated by computer programmes. Traditional approaches in combating spammers mostly lie on user features, such as the completeness of user profiles, the number of users’ activities, and the content of posted tweets etc. These methods may achieve good results when they are designed. However, spammers will evolve in order to survive. Hence the feature-based approaches will gradually lose their power. After careful analysis on a real SNS, we find that people will rebuild in cyber social networks their communities in the physical world. Interestingly, spammers also have to construct their communities in order to hide themselves, because separated users will be easily detected by anti-spam tools. Moreover, it is very hard for spammers to sneak into normal users’ communities since a community member needs to have links with most other members. Based on this observation, we propose a novel approach to judge a user by the features of both himself/herself and the accounts within the same communities as him/her.