Recommendation algorithm combining the user-based classified regression and the item-based filtering

Abstract

With the expansion of the Internet services, providing personalized product recommendations has become one of the most important ways to attract customers. Especially, collaborative recommender systems have achieved widespread success on the web. Information of products is recommended to the users based on their nearest "neighbors" who have similar interests. It is widely known that there is a sparsity problem in such systems. However, according to our research, there are other problems: one is that the typical collaborative algorithm loses some important parameter when it predicts the ratings, because there might be a strong similarity between the users who give very different ratings. Another is that the classification information of resources is not used. To solve these problems, we have proposed a recommendation algorithm combining the user-based classified regression and the item-based filtering. The experiment results show that performance is improved after applying the new algorithm.