Social Behaviour based Metrics to enhance Collaborative Filtering

Abstract

Expeditious growth of Internet and related network technologies has spectacularly increased the popularity of social networking systems such as blogs, forums, reviews sites etc. These systems allow the web users to share and disseminate their experiences and opinions with millions of users across the globe. This collaborative behavior of community can be observed as an electronic word of mouth (e-WOM) and can be utilized by the collaborative filtering systems to enhance the quality of recommendations. Despite of this importance very few studies have considered “social” aspect of user. This paper explores the role of explicit social relationship by presenting two novel similarity metrics. First metric is based on the social behavior (SB) that measures similarity between two users on the basis of “how similar they are in their social relationship”. The second metric integrates the (Hybrid) social similarity with the interest similarity between two users. The efficacy of proposed metrics has been evaluated over trust aware SFLA based collaborative filtering recommender system. Experimental study conducted on Epinions datasets indicate that for small set of target users, collaborative filtering (CF) system developed using social behavior metric performed better than Hybrid CF and conventional CF approach. However with the increase in percentage of active users, hybrid approach starts dominating and provides better recommendations.