Analysis of probabilistic models for influence ranking in social networks

Abstract

Influence is a phenomenon occurring in every social network. Network science literature on Influence ranking focuses on investigation and design of computational models for ranking of nodes by their influence and mapping the spread of their influence in the network. In addition to this contemporary literature seeks efficient and scalable influence ranking techniques that could be suitable for application on massive social networks. For this purpose joint and conditional probabilistic models could be a way forward as these models can be trained on data rapidly making them ideal for deployment on massive social networks. However identification of suitable predictors that may have a correlation with influence plays a major role in deciding the successful outcome for these models. The present investigation proceeds with the intuition that interaction is positively correlated with influence. Furthermore, through extensive experimentation it identifies a joint probabilistic model and trains it on interaction characteristics on nodes of a social network for influence ranking. A qualitative analysis of these models is presented to highlight its suitability.