Study On Collaborative Filtering Recommend Technology Based On The Interest Change Of User

In recent years, the E-commerce develops rapidly along with the popular ofcomputer and Internet, people online shopping become more and more convenient, onone hand the online trading greatly satisfy the shopping needs of people, but the otherhand the problem of information overload has been increasingly make troubles tocustomers. Facing the variety goods display in the online store, how to quickly find thesuitable goods is the urgent task of the modern electronic commerce site, thepersonalized recommendation system is just develop in this situation. It can recommendthe best satisfy and suitable goods to customers, have a good practical significance andbusiness prospects, and make more and more contribution to modern business as a newkind of application technology.Collaborative filtering technology is the most successful one among the severaltechnologies of the personalized recommendation, it include the cooperative filtertechnology based on the user and the cooperative filter technology based on theproject.This technology well received by users rely on the good algorithm thought andthe outstanding recommendation results. But in the application process, along with theincreasing of users and projects, the site structure and content become more and morecomplex, this technology appeared several problems such as data sparsity, user interestschange, cold-start etc. To solve these problems, many scholars already put forward lotsof significance solutions, but these problems are still serious and need more research upto now.This paper mainly studies the problems appeared in the application process ofcollaborative filtering technology, aim at the situation that it can’t provide the satisfyrecommendation when the interest of users has changed in the application of traditionalcooperative filter, analyzed its causes and put forward a new kind of algorithm ofcooperative filter based on the change of user’s interest.This algorithm fully consideredthat the importance of user scores will decay following the run of time, first clusteringthe users who have same features, then use the user data and forgotten function tocalculating the similarity of scores between the target users and the others which belongof the same kind of cluster,and sort the scores from high to low, screened out severalusers as the nearest neighbors of target users, and calculating the forecast scores aboutthe not scored project of target users through the combined with neighbor’s score and prediction scoring formula, finally choose the N projects which row in the top offorecasting to recommend to the user. This paper also puts forward the mode method tosolve the cold-start problem of collaborative filtering technology, it is through take thescore that several times appear in the scores of target users as the forecasting score ofnew project, even if there appear some new projects, this system also could makerecommendation to the users.This paper made a test based on the improved algorithm, the experimental resultsproved that this improvement of the algorithm is logical and effective, even if theinterest of users has changed, the recommend system also could give them a satisfactorypersonalized recommendation.