Ranking Influencers in Social Collaboration Networks

The emergence of social collaboration platforms has created new ways for people or groups to interact and work collaboratively to achieve mutual goals. The research community has highlighted the importance of measuring influence in such social collaboration networks to recognize key contributors and better facilitate collaboration. However, classic graph-based approaches often fail to elaborate on relations between people and the items they collaborate on, which is the essence of such social collaboration platforms. Moreover, with traditional methods it is usually difficult to incorporate auxiliary information about users and items into the modeling process. Therefore, in this paper we draw an analogy between recommender systems and social collaboration networks, and propose a method to harness information about users and items to model influence diffusion in collaboration networks. With this analogy, we quantify social influence by measuring the latent influence of each user-item pair under a unified framework based on factorization machines. Experiments conducted on a real-world dataset collected from GitHub attest the effectiveness of the proposed framework in providing better rankings for top influencers than several baseline methods. In addition, interesting analyses and discussions on the results are also provided in the paper.