A new dissimilarity measure for online social networks moderation

Abstract

Online Social Networks OSN and Virtual Communities VC software are vital tools useful to connect organizations with customers or community members. As these tools become more ubiquitous with general population, different managment problems start to arise. As these members' interactions become large, it is impossible for manual handling of moderation tasks without automatic or semi-automatic techniques.Of course, web mining techniques are very useful for understanding text patterns or browsing patterns in Websites, opening an opportunity to develop new algorithms to discover community members' patterns which have to be moderated. There have been previous work done on text patterns discovery for moderation that have reduced the moderation task to finding spammers. However, the moderation problem is much more complex: it involves not only text but also behavior patterns from bulling of other members to fights between them. We present a dissimilarity measure which includes human interaction aspects combined with these interactions' contents semantics not just free words of text. We show how we successfully applied our method into a real Virtual Community of Practice to detect users that should be moderated.