Research On User-adaptive Item-based Collaborative Filtering Algorithm

The mass of information generated by the rapid development of Internettechnology could easily lead to information overload, resulting in the problem thatpeople couldn’t quickly find required information from the massive information.Solving the information overload problem to help people get required informationquickly has attracted researchers’ wide attention, a variety of recommender system hasbeen proposed. Recommender system is a system that provides users with personalizedinformation service, whose core component is the recommendation algorithm.Collaborative filtering recommendation algorithm as one of the most successfulalgorithms currently use, become an important research direction in the field ofrecommender system. Although many improved algorithms are proposed on the basis ofcollaborative filtering algorithm, the recommendation quality is still not high.This paper on the basis of collaborative filtering algorithm and its improvedalgorithms, amends item similarity measure method and rating prediction method ofItem-based collaborative filtering algorithm, in order to more accurately predict therating of user to unknown item, thereby improving the recommendation quality.The main work includes:①Introduce the development process, composition structure, evaluation criterionand related technology of recommender system, focus on analyzing and summarizingthe principle, classification and problems of collaborative filtering recommendationtechnology.②Traditional Item-based collaborative filtering algorithm ignores the impact ofthe target user rating habit on recommendation effect when predicting the rating oftarget user to the unknown item, resulting in poor accuracy of rating prediction. Thispaper introduces the target user rating habit to rating prediction, and proposes animproved rating prediction method.③A user-adaptive Item-based collaborative filtering algorithm is proposed.Traditional Item-based collaborative filtering algorithm regards every rating of co-ratedusers (users rating both two items) as equal importance when calculating the similaritybetween items, and ignores the impact of the similarity between co-rated users andtarget user on the similarity between items, resulting in the similarity between items andtarget item become very high, but actually their similarity is low in the eye of target user. This paper’s proposed algorithm uses the similarity between co-rated users and targetuser as the rating importance weight of co-rated users, in order to adaptively measurethe similarity between items.④Use the dataset provided by MovieLens site and carry out simulationexperiment on Matlab platform, through comparing with the traditional Item-basedcollaborative filtering algorithm and other similar improved algorithms to verify theeffectiveness of the proposed algorithm.The experimental results show that comparing with traditional Item-basedcollaborative filtering algorithm and other similar improved algorithms, our algorithmhas a higher rating prediction accuracy rate, thereby improving the recommendationquality.