Towards the design of a platform for abuse detection in OSNs using multimedial data analysis

Online social networks (OSNs) are becoming increasingly popular every day. The vast amount of data created by users and their actions yields interesting opportunities, both socially and economically. Unfortunately, these online communities are prone to abuse and inappropriate behaviour such as cyber bullying. For victims, this kind of behaviour can lead to depression and other severe problems. However, due to the huge amount of users and data it is impossible to manually check all content posted on the social network. We propose a pluggable architecture with reusable components, able to quickly detect harmful content. The platform uses text-, image-, audio- and video-based analysis modules to detect inappropriate content or high risk behaviour. Domain services aggregate this data and flag user profiles if necessary. Social network moderators need only check the validity of the flagged profiles. This paper reports upon key requirements of the platform, the architectural components and important challenges.