Mihaylov, Todor, Georgi Georgiev, and Preslav Nakov. "Finding opinion manipulation trolls in news community forums." Proceedings of the nineteenth conference on computational natural language learning. 2015.

This study addresses the challenge of automatically identifying opinion manipulation trolls in user forums of electronic news media. Given the difficulty in defining trolls and obtaining training and testing data, the research pragmatically assumes that users labeled as trolls by several others are likely trolls. Different variations of this definition are experimented with, and a classifier is trained using a rich feature set. The results demonstrate that the classifier can distinguish likely trolls from non-trolls with high accuracy, ranging from 82% to 95%.