Characterization and Detection of Political Manipulation in Online Communities

Abstract

As an increasing number of people share their opinions in online communities, attempts to manipulate these opinions are also rising steadily. In particular, when manipulation is used for political purposes (e.g., to decrease the credibility of a presidential candidate), it can change the outcomes of elections, thus having a long-lasting, negative impact. This work is aimed at detecting such political manipulation. We collect 377K opinions from Internet forums at the peak of a presidential campaign in South Korea. We find various characteristics of political manipulation, including the distortion of opinion polls with numerous IDs and the continuity of the same, strong political inclination throughout manipulative posts. Based on these characteristics, we implement a detection system and evaluate it with the collected data. We demonstrate that this system accurately discovers more than 90% of online accounts involved in political manipulation. We also find clues indicating that greater than 10% of these accounts are collectively controlled by a few manipulative users. We believe that the proposed system can ensure people read trustworthy opinions and also relieve web administrators from manually screening for manipulative activities.