Social News Website Moderation through Semi-supervised Troll User Filtering

摘要: Recently, Internet is changing to a more social space in which all users can provide their contributions and opinions others via websites, networks or blogs. Accordingly, content generation within webs has also evolved. Users of news sites make public links stories, so that every user comment them other users’ comments related the stories. In these sites, classifying depending on how they behave, be useful for web profiling, moderation, etc. this paper, we propose new method filtering trolling users. To end, extract several features from profiles order predict whether troll not. These are used train machine learning techniques. Since number very high labelling process laborious, use semi-supervised approach known as collective reduce efforts supervised approaches. We validate our with data ‘Meneame’, popular Spanish site, showing achieve accuracy rates whilst minimising task.