Research On Several Issues In Collaborative Filtering Technology

With the rapid development and wide application of Internet technology, thedesire of people to obtain information can be fully met. How to get useful part fromthe growing mass of information becomes the bottleneck of the effective use ofinformation. The personalized recommendation system is one of the effective meansto solve the problem. Recommendation algorithm is the key technology ofpersonalized recommendation system, and determines the pros and cons of therecommendation system. Collaborative filtering algorithm is one of the mostsuccessful recommendation algorithms. It seeks similar neighbor-users according tosome strategies based on the user’s behavioral characteristics and makesrecommendation according to the neighbor-users’ interest finally. However,collaborative filtering algorithm is also facing some problems with the increasing ofcontent and users in recommendation system and the increasing user requirements onthe recommendation qualityThe information of user behavior characteristics is still very scarce in therecommendation system. Data sparseness problem is a major factor affecting therecommendation quality. In order to improve the recommendation quality in the caseof data sparseness, an improved collaborative filtering algorithm was proposed. Usethe SOFT-IMPUTE algorithm to fill sparse rating matrix. Afterwardsuser-similarities and their confidence factor were calculated using the completefilling matrix. Utimately, the recommendation forecast was made. Comparativeexperiments on typical dataset show that the algorithm is able to achieve betterresults even with extremely sparse data.In addition, the information in the Internet cost low and accessed convenient. Itis necessary to mine long tail information in recommendation system. To improvethe recommendation systems ability of mining unpopular items, an improvedcollaborative filtering algorithm is proposed. Based on traditional algorithm, items’popularities is considered as weighting factor in similarity calculating andrecommendation process, to boost reliability of user-similarities calculating andinfluence of unpopular items in finally recommending. Comparative experiments on typical dataset show that the algorithm is able to mine unpopular items effectivelyunder the premise that maintaining or even improving recommendation accuracy.