Mining Social Information For Recommender System

With the development of e-commerce and mobile commerce, recommender system has been more and more important in both the research field and practice field. Traditional recommender system researchers make use of user rating lists as the basis of recommendation, with ignoring other factors which affect users’ preferences. However, in fact, user preference and purchase decisions depend on both themselves experience and social information. Previous recommendation researches take social information into considerations when they estimate user preferences and user ratings, but they only focus on user behaviors which may bring the privacy problems or just make the distance between users to measure the strength of social influence between users. This research focuses on how to make full use of social information to make more accurate recommendation. Social information refers to information from social environment, which not only includes information from close friends of users, but also includes information from public comments. According to the theoretical foundations learnt from the social researches, behavior researches and recommendation researches, we propose three novel social recommendation methods. Field experiments described in this research compare the accuracy between existing recommendation methods (including the classical traditional recommendation methods, recent classical recommendation methods and previous social recommendation method), social recommendation method considering social information mined from friends, social recommendation method considering social information mined from public comments and social recommendation method considering social information mined from both friends and public comments. Experiment results show our proposed three social recommendation methods could achieve much higher performance on precision, recall and F-measure than previous recommendation methods. Especially when users do not provide any ratings to movies, the social recommendation methods even become irreplaceable as other tested recommendation methods could not work at all on this condition. We also find the social information mined from friends play a more important role than the social information mined from public comments for the estimation of user preferences and the estimation of t ratings.