Anomalous User Comment Detection in Social News Websites

The Web has evolved over the years and, now, not only the administrators of a site generate content. Users of a website can express themselves showing their feelings or opinions. This fact has led to negative side effects: sometimes the content generated is inappropriate. Frequently, this content is authored by troll users who deliberately seek controversy. In this paper we propose a new method to detect trolling comments in social news websites. To this end, we extract a combination of statistical, syntactic and opinion features from the user comments. Since this troll phenomenon is quite common in the web, we propose a novel experimental setup for our anomaly detection method: considering troll comments as base model (normal behaviour: ‘normality’). We evaluate our approach with data from ‘Menéame’, a popular Spanish social news site, showing that our method can obtain high rates whilst minimising the labelling task.