A peer-based social relationship enhanced recommendation model

Recommender systems have been developed to address the “information overload”issue by providing users with potentially useful products or services. While mostof the current systems are continuing to deal with globally collected large numberof users and items, little attention is paid to the situations where users ask forrecommendations through a limited number of personal social circles. Social networks, on the other hand, are one of the most popular channels throughwhich people share and exchange information. Many studies in recent years haveshown that incorporating social relations and interactions into recommender systems will significantly improve recommendation qualities. To make superior recommendationsfrom peers, we are especially interested in developing a recommendationmodel that emulates the natural recommendation style in real life. In this thesis, we propose a peer-based social recommendation model that imitates the natural recommendation process in social networks. The model forms neighborhoods from peers across social circles, through which the peers participatein the recommendation process by propagating requests and responses in arelay fashion. Furthermore, we develop a machine learning method to measure tiestrengths among the peers in a social network, based on various social relationships.Generally speaking, the stronger the tie strength between two individuals,the more similar interests they may commonly share. Furthermore, the learnedtie strengths are then incorporated in recommendation process to increase theaccuracy and relevance of the recommendations. We have conducted comprehensive experiments by using the real datasets frompopular social media services. The evaluation results demonstrate that our proposed recommendation model outperforms other popular and state-of-the-art recommendationmethods in terms of widely accepted evaluation metrics.