Recommendation Based on Influence Sets

Recommender systems have been successful in several domains, such as E-Commerce and Web personalization. But the traditional user-based collaborative filtering (CF) approaches existing for building recommender systems have shown some fundamental problems, such as sparsity and scalability. Recently, item-based CF algorithms have been presented to deal with the scalability problems associated with user-based CF approaches. However, item-based CF algorithms still suffer from the data sparsity problems. This paper presents a different view of integrating Influence Sets into the recommendation process, which is a hot topic in information retrieval system. RIS, a novel item-based CF approach, combines the effects of k nearest neighbors with reverse k′ nearest neighbors for a target item to enhance the density of information. Moreover, new prediction generation methods are defined for this new recommendation mechanism. Our experimental results show that RIS can achieve better prediction accuracies than traditional item-based CF algorithm and alleviate the dataset sparsity problem effectively.