The Research Of The Collaborative Filtering Recommendation Algorithm Based On Time Weighted And Score Predicted

We are getting into the era of information overload from the era of information shortage gradually with the development of information technology rapidly. It is increasingly difficult to get information of useful and interesting from massive information. Recommendation system as an important tool to overcome information overload has captured the attention of industry and academic consistently. Collaborative filtering algorithms play an important role in the recommendation system. Therefore, it has more attention in application and scientific research. Conventional collaborative filtering algorithm does not consider recommendation system’s dynamic characteristics and not adapt to the change of user’s interest. It also affected by more and more sparse user rating data. So, the performance of recommendation has declined continuously. However, although recommendation algorithms of existed about improvement of collaborative filtering algorithm can adapt to the change of user’s interest. There is just a slight improvement of the performance of recommendation system, because it is also affected by sparse user rating data seriously. In order to improve the accuracy of similarity between users and users or items and items. And it also can adapt to the change of user’s interest based on collaborative filtering algorithm. We propose a novel method which combines the score prediction with traditional collaborative filtering based on time weighted. At first, we study the algorithms of score prediction of existed and decide to exploit an appropriate and effective score prediction algorithm to get the rating data. Then, we give a part of higher rating data with time value of a reasonable hypothesis. Finally, we will calculate the similarity of user or item by the time weighted method on new score matrix. We compare our proposed method with conventional algorithm on MovieLens dataset. The experiment result is shown that the proposed method has declined injurious influence of sparse data to similarity and proves it can improve the accuracy of similarity and the quality of recommendation.