How to predict social trends by mining user sentiments

Abstract

The majority of techniques in socio-behavioral modelling tend to consider user-generated content in a bulk, which may ignore personal contributions of specific users to predictability of the system. We propose a novel user-based approach designed specifically to capture most predictive hidden variables which can be discovered in a context of specific individual only. User content is assessed to determine both the subset of best, “expert”, users able to reflect particular social trend of interest, and their transformation into feature space used for modelling. The technique is tested on a case study of Chicago crime rate trend prediction using historical tweets of selected citizens. We also propose a new user ranking approach which exploits the concept of user credibility.