Recommendation System Based On The Trust Network Research

Information grows explosively as a result of the rapid development of Internet and E-Commerce, could be regarded as information overload. Due to that, recommender systems come into being. Collaborative filtering is one of the most successful technology in recommender system, which recommends items on basis of the rating matrix. However, the rating matrix is usually very sparse, user similarity couldn’t be computed for lacking of common rated items between users. This may result in unsatisfactory recommendations.Users interact more frequently from web1.0to web2.0era. Some E-Commerce sites set up based on web2.0, users could rate or review items to express their opinions. On one hand, system could have better understanding of users’attitudes; on the other hand, users communicate more often. Therefore, system provides the evaluation between users, user could build trust relationship with the user whose reviews are valuable and real. The trust-aware network forms.This paper combines trust-aware network with recommender system, replacing users’ similarity with trust. Whereas the trust matrix may be very sparse, we should predict new trust relationship to extend the trust-aware network. This paper proposes Extended Trust-aware Network based Recommendation Framework (ETNR). With regard to binary trust-aware network:users may be similar because users trust some other users simultaneously, based on this proposes GenTrust algorithm to extend trust-aware network; since the trust value in binary trust-aware network could only be zero or one, proposes IndegreeTrust algorithm to differentiate the users trusted by the same user. Toward non-binary trust-aware network:adopt the widely applied Slope One algorithm to extend the trust-aware network. The results of experiments demonstrate that there is improvement in accuracy compared to some other algorithms. The recommendation list produced by users’similarity is more novel, while list generated by users’trust is more reliable. This paper proposes CombineList algorithm to combine these two recommendation list, considering user’s different demand for novelty and reliability.