Research On Collaborative Filtering Algorithm In Recommendation System

With the rapid development of Internet, the information online is getting more and more large, Internet is become an indispensable part in people's life.Because of the large information, it needs too much time for people to search for the information or shopping online. In order to help people find the information or products more expediently, recommendation system was suggested. Recommendation system can be divided into many types, the most important and widely used one is the collaborative filtering recommendation system. The goal of collaborative filtering system is to suggest new items or to predict the utility of items for users based on the users'previous likings and the opinions of the other like-minded users.At present, the collaborative filtering technology is a hot topic, and the researches and applications in the field have achieved great success,but there are still many problems to be resolved.In the application of the collaborative filtering recommendation system, there are four main important challenges.The first challenge is the accuracy problem.As the number of the products online is too large, but:the evaluation of users is limited, making the data about users and products is very sparse, so the accuracy of the system is affected.The second challendge is the scalability problem. How to make sure the system adapt to the growing number of the users and products, is a big problem. The third challendge is the "cold start" problem. Since the collaborative filtering is based on the historical record, so the system cannot recommend items to him or recommend new items to users.The fourth challendge is the real-time problem. Now that many recommendations are based on line, which requires the high real-time feature. Many online recommendations achieve the requirement but lose the accuracy, so how to resolve the issue is another problem.As the accuracy of the algorithm is the most import issue, there are many researches on the issue. This paper is carried out in order to resolve the issue, and makes the main contributions as follows:First, this paper compares the several methods of processing no-rated items in collaborative filtering algorithm. As the extreme sparsity of data, the users-items matrix contains a large number of no-rated items, and different method of processing no-rated items affects the accuracy of the system. Among the current algorithms of calculate the similarity, the experimental results show that setting the value of no-rated item as 0 can achieve the highest accuracy.Second, this paper proposed an optiomized item-based collaborative filtering algorithm. The traditional item-based is based on such an assumption:If the majority of users rate some items similarly, the target user will rate the items similarly. While calculating the similarity of two items,we obtain the ratio of users who rated items to those who rated each of them. The ratio is taken into account in this method. The experimental results show that the proposed algorithm can improve the quality of collaborative filtering.