Collaborative Filtering Algorithms For Personalized Recommender System

With the continuous development of Internet technology,network information has ac-cumulated over time,bringing about issues such as abundance information,information with poor quality and low value.Recommendation system is one of the effective techniques for solving these issues.Recommendation system can discover interests and hobbies of users through their historical behavior data,and accurately recommend items in which users are interested.Currently,collaborative filtering(CF),one of the common techniques for recommen-dation systems,deals with historical behavior data of users.However,the behavior data has issues such as data sparsity and rating diversity,which would have a negative impact on recommendation performance.In this thesis,a clustering algorithm in machine learning and an assumption of truth discovery are introduced to collaborative filtering methods when handling the behavior data.In doing so,the issues of data sparsity and rating diversity can be solved partly,and thus the recommendation performance can be improved.The innovation work of this thesis is summarized as follows.(1)In order to reduce the impact of data sparsity on the recommendation performance,this thesis proposes a Jaccard coefficient-based bi-clustering and fusion(JC-BiFu)user collaborative filtering method.JC-BiFu clusters users according to similarity by intro-ducing a density peak clustering algorithm.Through calculating the Jaccard similarity coefficient between users,the unrated items of users are predicted and complemented,which solves the issue of data sparsity and improves the accuracy of recommendation.Different from other similarity measurements,the Jaccard coefficient measures the sim-ilarity between users by the rate of the number of common rating to the total number of rating,which decreases the sensitivity of JC-BiFu to the rating values.We perform a large number of simulations on three real movie data sets,and verify the effectiveness of JC-BiFu.(2)In real life,users often have subjective emotions when they make feedback on items.In order to avoid the impact of subjective emotions,this thesis introduces the assumption of truth discovery.The assumption is that each item has its truth value in theory,which requires to be found to measure the true level of the item.Based on this assumption and JC-BiFu,this thesis proposes a user collaborative filtering algorithm,called truth discovery-based user collaborative filtering(TDUCF)algorithm.TDUCF can solve the issues of data sparsity and rating diversity in the same time by predicting the missing values in the rating matrix and finding the reliable users according to the truth of item-s.Experimental results show that TDUCF has a better recommendation performance compared to the state-of-the-art methods.(3)According to the above two collaborative filtering methods,this thesis develops a movie recommendation system which would record the historical ratings and the recommen-dation results for a user once he logins in.This system can automatically generate a new movie collection that is more in line with the hobby according to the rating of users on the given movie list.In addition,by collecting the rating of the current popular movie on the main interface,we can have more accurate recommendations to users.