Collaborative topic regression for predicting topic-based social influence

Abstract

The rapid growth of social networks and their strong presence in our lives have attracted many researchers in social networks analysis. Users of social networks spread their opinions, get involved in discussions, and consequently, influence each other. However, the level of influence of different users is not the same. It varies not only among users, but also for one user across different topics. The structure of social networks and user-generated content can reveal immense information about users and their topic-based influence. Although many studies have considered measuring global user influence, measuring and estimating topic-based user influence has been under-explored. In this paper, we propose a collaborative topic-based social influence model that incorporates both network structure and user-generated content for topic-based influence measurement and prediction. We predict topic-based user influence on unobserved topics, based on observed topic-based user influence through their generated contents and activities in social networks. We perform experimental analysis on Twitter data, and show that our model outperforms state-of-the-art approaches on recall, accuracy, precision, and F-score for predicting topic-based user influence.