Font Size: a A A

Research On Collaborative Filtering Recommendation Integrating User Trust And Tags

Posted on:2020-08-22Degree:MasterType:Thesis
Country:ChinaCandidate:H WuFull Text:PDF
GTID:2428330596468143Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
The advent of the era of big data has made information begin to flood,and the problem of "information overload" has become more serious.The recommendation technique based on collaborative filtering algorithm provides a good help to solve this kind of problem.However,the collaborative filtering algorithm still has the problems of “data sparse” and “cold start” inherent in the recommendation system,which seriously restricts the development of the recommendation system.In order to solve this problem,more and more auxiliary information has been introduced into the recommendation system.People can provide users with personalized needs more accurately through auxiliary information.Among these supplementary information,social trust and socialization tags are particularly noticeable.This paper proposes corresponding improved algorithms based on these two kinds of auxiliary information.This paper mainly introduces three aspects: collaborative filtering algorithm incorporating potential trust model,matrix factorization based on user trust relationship,and tag-based probability matrix factorization.The main research contents are as follows:(1)In view of the problem of insufficient social information,this paper proposes a potential social trust model,which constructs a trust model from four aspects: global trust value and expert system,improved trust propagation,and improved Pearson coefficient.The core idea of the algorithm is to propose the idea of populating the trust matrix on the basis of the existing trust propagation,in order to enhance the richness of the social network.Through the experimental verification on the FilmTrust real data set,it is proved that the algorithm can enrich the social network to a certain extent,thus improving the accuracy of recommendation.(2)In the trust-based matrix factorization algorithm,the implicit feedback algorithm(TrustSVD)that combines explicit trust relationship scores is a better algorithm for the matrix factorization class.Based on the TrustSVD algorithm,this paper proposes a more accurate trust-based matrix factorization: TrustSVD++ algorithm.Based on the TrustSVD algorithm,we incorporate implicit feedback from similar users and linearly combine with the implicit feedback of trusted users.At the same time,we limit the range of trusted users in the TrustSVD algorithm,so that the algorithm can bring more accurate hidden.In the end,we add two factors in the trust decomposition: the impact of the inherent trust factor of the user on the trust prediction,and the implicit feedback that trust users bring to trust prediction.Experiments on the two datasets,FilmTrust and Ciao,show that the TrustSVD++ algorithm can improve the recommendation to a certain extent.(3)The common tag similarity calculation lacks the combination of the user's score,and we know that the user's score actually shows the user's interest.Therefore,this paper proposes a tag similarity combining user score and tag frequency,and then integrates the influence factors of user tags and project tags on the basis of probability matrix factorization(PMF),and proposes a tag-based probability matrix factorization algorithm.In order to verify the effectiveness of the algorithm,the similarity of the tags commonly used in the paper was compared.Experiments in two real data sets show that the proposed algorithm can improve the recommendation effect to some extent.
Keywords/Search Tags:collaborative filtering, recommendation system, social trust, socialization tag, trust model, matrix factorization, probability matrix factorization
PDF Full Text Request
Related items