Using compression models for filtering troll comments

Internet is evolving. How the content is generated has changed and currently, users and readers of a site can create content. They can express themselves showing their feelings or opinions commenting diverse stories or other users´ comments in social news websites. This fact has led to negative side effects: the appearance of troll users and their contents seeking deliberately controversy. In this paper we propose a new method to filter trolling comments using compression models. Normally, Vector Space Model representation use is quite common but these filters can be attacked. To this end, we validate our approach with data from ‘Menéame’, a popular Spanish social news site, training several compression models, showing that our method can maintain high accuracy rates whilst making such filters difficult to defeat.