The Research Of Time Weight Distributed Collaborative Filtering System

With the emergence of the information overloading, an engine which could effectively solute information overloading and provide the users with their required information is urgently needed. Thus, personalized recommendation technology came into being. The recommendation system recommended for users to meet their information needs and interests by analyzing users’ behavior history, a good recommendation system is a system that provides information to meet the users’ interests in a short term. While the amount of data continues to grow exponentially, the traditional collaborative filtering algorithm faced with challenges about accuracy, scalability, cold start and others.In this paper, we proposed the time weighted distributed collaborative filtering system. The system is optimized mainly in two aspects. One is an algorithm optimization; the other is a speed optimization. The system’s process is divided into three parts: part of the users’ login predictor, part of recommender, part of personalized recommendation evaluator.First, the part of the users’ login predictor predicts users’ login status on a given day based on the analysis of users’ historical login data. The users will be divided into two types: the users who will login on that day and the users who won’t login on that day. The system makes personalized recommendations for the users who are predicted to login, and recommends popular items to the users who are predicted not to login.Second, the part of recommender is divided into personalized recommendation and popular items recommendation. Personalized recommendation is implemented by the item-based collaborative filtering recommendation; the recommendation process is composed of two similarities’ weighted fusion similarity calculation, ratings’ time decaying and generating a recommendation list. In the popular items recommendation, the items are sorted by popularity and recommend the most popular items to the users directly.Third, we applied the time weighted distributed collaborative filtering system to a real scene. We started from business needs of movie recommendation system, analyzed the system performance with relevant data, and evaluated the performance of the system finally. We get the conclusion that the accuracy and speed of system are improved.