A Personalized Recommendation Algorithm on Integration of Item Semantic Similarity and Item Rating Similarity

Abstract

With the rapid development of the Internet and the wide application of e-commerce, recommender system has become a necessity and collaborative filtering is the most successful technology for building recommendation systems. There are many problems in the recommendation approaches, such as data sparsity problem, the issue of new items and scalability issues. Item-based collaborative filtering algorithms can improve the scalability and the traditional user-based collaborative filtering methods, to avoid the bottlenecks of computing users’ correlations by considering the relationships among items. But it still worked poor in solving the issues of sparsity, predictions for new items. In order to effectively solve several problems, this paper presented a recommendation algorithm on integration of item semantic similarity and item rating similarity. The item semantic similarity is calculated combining Earth Mover's Distance and Proportional Transportation Distance, which can utilize the semantic information to measure the similarity between two items based on a solution to the transportation problem from linear optimization1. Then producing recommendation used item-based collaborative filtering integrating the semantic similarity and rating similarity. The presented approach can effectively alleviate the sparsity problem in e-commerce recommender systems.