A Community Based Social Recommender System for Individuals & Groups

Abstract

Social networks provide rich new types of data that promise to personalize and improve Recommender Systems. One promising type of data extracted from social networks is relationships. While existing recommenders have capitalized on the use of explicit relationship data such as friendships, the use of implicit relationship data (that gathered from less obvious connections such as co-occurrence) offers great potential for diversifying recommendation for both individuals and groups. Our novel Community Based Social Recommender System (CBSRS) utilizes this new social data to provide personalized recommendations based on communities constructed from the users' social interaction history with the items in the target domain. It allows quick and efficient recommendations to groups as well as individuals. We propose and evaluate such approach using the Internet Movie database (IMDb). We use the underlying social network graph of the movies based on their common reviewers to model the generic network of interests. Communities are then discovered and used as a basis to provide extensive and diverse recommendation for one user, couples or group of users (friends, family, co-workers) offering them movies of their common interests. Finally we demonstrate that the proposed CBSRS increases the accuracy in the results whilst it addresses the cold sparsity of data and start problem for new users and items and provides recommendation for both individuals and groups.