Discovering Audience Groups and Group-Specific Influencers

Abstract

Recently, user influence in social networks has been studied extensively. Many applications related to social influence depend on quantifying influence and finding the most influential users of a social network. Most existing work studies the global influence of users, i.e. the aggregated influence that a user has on the entire network. It is often overlooked that users may be significantly more influential to some audience groups than others. In this paper, we propose AudClus , a method to detect audience groups and identify group-specific influencers simultaneously. With extensive experiments on real data, we show that AudClus is effective in both the task of detecting audience groups and the task of identifying influencers of audience groups. We further show that AudClus makes possible for insightful observations on the relation between audience groups and influencers. The proposed method leads to various applications in areas such as viral marketing, expert finding, and data visualization.