Recognition of manipulated posts based on SVM classification on bulletin board system

Abstract

Inspired by the fact that online Public Relations (online PR) companies manipulate the online information and confuse people about the truth of information, a novel problem of identifying messages that are controlled by online PR companies is presented. Combined with the knowledge of characteristics of opinion leaders, methods of agenda setting and strategy patterns of online PR companies, a set of features which can identify manipulated posts on bulletin board system (BBS) is proposed and verified by using classification methods. Experiments with data from real-world BBS are conducted to evaluate the ability of the feature set. The verification using Support Vector Machines proves that the feature set can be used for identification with the accuracy surpassing 74%.