Research Of Collaborative Filtering Recommender Algorithm Based On Time Weight

Following with the fast-development of modern science and technology, computer network has been used in all walks of life widely. The electronic commerce system has also been associated with each of us closely, which makes our life more convenient and quick. At the same time, large and complex information of electronic commerce systems can affect our appropriate choice of goods and services. Therefore, the research of recommendation algorithm arises at the historic moment. Among them, the collaborative filtering algorithm as the darling of the recommended areas is widely concerned. At present, as the time for big data era is coming, the update speed of information become faster and faster, thus, the problem of out-of-date information become one of the key factors influencing the accuracy of recommendation algorithm.As is known to all, the user’s interest is changing over time. The traditional collaborative filtering algorithm ignores the influence of this change, and can’t make good recommendation for users. To this end, this paper introduced the concept of information retention period inspired by the information half-value period, namely a time window of information influence remains the same basically. At the stages of nearest neighbor searching and predictive scoring, we use a novel time weighted function to put time weight to the user’s score, and make the recently scores gain greater weight to show higher reference value for predicted ratings. We called it an improved collaborative filtering algorithm based on time weight (NTWCF), it can greatly improve the accuracy of predicted ratings.On the other hand, with the increasing of the number of users and items in electronic commerce systems, the real time requirement of recommendation systems is more and more prominent. This paper proposes time weighted item clustering algorithm to optimize NTWCF, and puts forward a collaborative filtering recommendation algorithm synthesizing time weight and item clustering (TWICCF). It improves the accuracy of algorithm as well as ensures the real-time response speed of recommendation systems. TWICCF uses K-means to cluster the items based on time weighted scores, and then searches the nearest neighbors of the target item in the set of some clusters. It can shorten the execution time of the algorithm to guarantee the real-time requirements effectively, and improve the accuracy of the algorithm to some extent.