Combining Content Prediction And The Item-based Collaborative Filtering Recommendation System

Recently, recommendation systems, especially collaborative filtering recommendation systems, have achieved widespread successes on the Web. However, the tremendous growth in the amount and variety of available information poses some austere challenges to recommendation systems, the problems of cold-start, sparsity and scalability in collaborative filtering recommendation are in dire need to be solved. To address these issues, the personalized recommendation systems and some recommendation methods are introduced in this paper, and the reasons of the problems of cold-start and sparsity in collaborative filtering recommendation are analyzed. After the research on the characteristics of item-based collaborative filtering recommendation and content-based recommendation, a hybrid recommendation method improving the item-based collaborative filtering through content-based recommendation is presented. The new method improves the recommendation of new items of the collaborative filtering using the analysis of similar items of content-based recommendation, and improves the shortage of extensibility in the. traditional collaborative filtering recommendation by replacing it with item-based collaborative filtering. In the hybrid algorithm, firstly, the characteristics of attributes of items are analyzed and the similar items to the new item are found by content-based recommendation. Then, the rate to the new item is forecasted by the rate to these similar items. Finally, the neighbor users on the similar items are computed by item-based collaborative filtering recommendation and the final rate of forecast is presented. So the problems of cold-start, sparsity and scalability are solved.Finally, a document recommendation system using the method of hybrid recommendation is shown, with its structure and function introduced in detail. The experiments taking on this system show that the method it used can not only efficiently solve the problems of cold-start, sparsity in traditional recommendation systems, but also has a faster responding speed compared with the traditional ones. Inthe meanwhile, the veracity of the new system is also comparable with the traditionalones.