Research On Collaborative Filtering Algorithm For Internet Recommender System

Usually, users have to face a mountain of information on their own in many internet applications. No matter familiar or not with this mass information, people hope to get useful and personalization information from that as quickly as they can. Recommender system has been brought forward to handle this situation. Based the historical information of users, it can make recommendation which best meet their desire and filter the information that useless to them One of the most typical internet applications of recommender system is E-commerce, in which recommender system play a role of sales to give advice and help people make better choice.Recommendation algorithm is the most important part of recommender system, which predicts users' rating for items or recommends good item to user according to users'profile, historical information and also the profile of items. Among all the recommendation algorithms, collaborative filtering is the most popular one. Although there are a lot of successful applications of recommender system in internet, however, with the development of web technology and the changing of user demand, collaborative filtering is facing these two issues:1) Incorporate with the profile of user and item to enhance recommendation precision. The profile of user usually refers to human demography information, while the profile of item includes shape and visit time and so on. 2) Users preference is always changing as time shifting. If the algorithm fails to follow the step of users, it could not make correct recommendations. So how to capture the changing preference need to be studied.In order to handle the problems mentioned above, we propose a time weight based collaborative filtering algorithm This algorithm introduces time weight to improve the recommendation compute process based traditional neighborhood-based collaborative filtering algorithm The time weight function can be constructing with the visit time of item, which maintain in the item profile. Intuitively, the changing of user's preference is related with time, so the time weight can measure this change to some extent. To bring it forward, a exponential time weight based collaborative filtering algorithm is proposed, also a logarithm time weight based one. Using precision as evaluation metric, a series of experiments based binary data set are designed to evaluate these two algorithms. According to the experimental result, logarithm time weight is blending well with user-based collaborative filtering algorithm and performs best. Comparing with the traditional neighborhood-based collaborative filtering, logarithm time weight based one make a little improvement in precision.