Action-Aware Restricted Stream Influence Maximization Model to Identify Social Influencers

The problem of influencer identification is an important problem in social network analysis, due to the impact of influential users one the opinions of their audience. Most of the existing approaches to identify influencers are developed for static networks, whereas the social networks are time-sensitive and evolving over time. Therefore, identifying influencers over a dynamic, or stream, social network is more adequate for such problem. However, the amount of work proposed for dynamic networks are limited. Recent work proposed that identifying influencers with respect to some analysis-specific restrictions (e.g. influencers' locations or Influence context) produces a more concrete analyses. Current models proposed to identify influencers are based on capturing the number of social actions triggered by an influencer's social action. These models do not differentiate between social actions' types and treat them indistinguishably. However, the type of an action a user select to do captures an important clues in how as user is influenced. In this paper we propose to solve Action-Aware Restricted Stream Influence Maximization (AR-SIM) problem that identifies the most influential social network users in real-time. We extend the Action-based dynamic model [5] to incorporate actions' types into the model. The model does not only differentiate between the actions' types, it gives the option to weight these actions differently; facilitating new approaches to identify influencers. We run the model with respect to a given set of commonly used restrictions. We adopted a sliding window to update efficiently the model in real time. The model is generic and can be used with any social network platform, actions types, and restrictions. We run our experiments using Twitter data where we differentiate between four action types: (tweet, retweet, reply and quote tweet) and with respect to location, topic and/or language restrictions. Our results shows that our new model is able to identify significantly different influencers based on the given actions wights. This should open the gate for more sophisticate and deeper understanding for influencers impact types over the social network. The model is generic and can be used in any type of social network.