New Methods for Ranking Influence in Social Networks

Abstract

In this work, propagation dynamics on social networks are studied in order to identify the most influential users. For this purpose, diffusion data has been collected during 4 weeks from a microblogging OSN (online social network) called Tumblr. Then, the propagation graph has been built and studied using the first 2 weeks data (period $T_1$). Subsequently, this graph has been used to predict th influencers during the last 2 weeks (period $T_2$ ). A ranking of influential nodes is obtained for $T_2$ , set as the ground truth. The aim is to predict this ranking using the data from $T_1$. Based on the average spread of users’ posts, rankings obtained with several techniques are tested and compared. These techniques include classical centrality measures used in the literature, the $T_1$ ranking itself, and new alternatives based on effective degree using local (network) information. Whilst all methods perform similarly when considering whole global ranking, differences among them appear when ranking the top influencers. For those, in general, the methods proposed here outperform the classical centrality measures.