Mining Spam Accounts with User Influence

Abstract

As the increasing development of online social networks (OSNs), spammers' attentions have been attracted from the traditional email field. Nowadays, advertisements, deception messages, illegal contents are prevalent in all kinds of ONSs. They're propagated from one to another arbitrarily, polluting the Internet environment, and what's more, resulting in a great many of security problems. Some previous works have been proposed to detect spammers according to user properties. The problem is that in order to prevent from being detected, spammers are likely to pretend to be normal, and what's more, some normal users also engage into spam spreading for financial benefits, making detection more difficult. In this paper, we solve the detection problem from the view of user influence. The basic of our work is that since spammers pretend to be normal, their influences should keep step with their normal behaviors. But when a spam campaign is launched, usually in order to influent others, a great many of spammers engaged into propagation, the original poster's influence would get a sudden increase, making him outstanding from the others. In this way, we can distinguish the original spammers and supervise from the root of the propagation tree. Our work is experimented on real data gathered from Weibo and shows inspiring results.