A Novel Approach to Trust based Identification of Leaders in Social Networks

Abstract

Recommender Systems has been a research hotspot in recent times as an efficient information filtering tool, to filter out useful required information from ever expanding web. The characteristics of social networks play a very important role towards behavioralmodeling of a trust network based Recommender System (RS). Similar to real world, in a social network also, it is important to objectively identify a member with high reputation who is heavily trusted by many members and hence his suggestions and inputs are most trust worthy for the whole community; that is, objective identification of leader is extremely important. In this paper, we propose a method to objectively identify leaders in a social network, we introduce new terms: Leadership score, prominence trust, peer inclusive factor, trust spread factor, trust maturity factor and trust penetration factors to portray a member more appropriately in a social network. We have calculated the leadership score using the real world dataset. Leadership score is taken as a linear additive function of prominence trust, engagement trust and peer inclusive factors with different weightage values. In our experiment, we have considered that leaders not only are followed by others, they are also socially engaged from their end to others in the social network. Prominence trust is a detailed characterization to get more accurate value of trust. Along with overall trust score, it is imperative to analyze more attributes to derive a more objective interpretation of the same. These attributes take care of trust score/reputation over period of time, number of trusting members, peer level interactions and also absolute score. To validate our leadership model in social network, we remove top 5%, 10% and top 20% of the members with high ‘leadership score’ and findout the reduction in the overall interactions. As further improvements to this work, a social network to be built with users and items and collect the data using positive and negative responses; use multiprogramming concepts to determine the optimum function for leadership score. This will aid towards further analysis and fine tuning of the models.