Trust Network Based Collaborative Filtering Recommendation Algorithm

Recommendation plays an important role in addressing on information overload, whichcould effectively help users to find the interested contents from magnanimous informationand has widely applied to many areas, such as news, books, movies, music and others.Currently, although a number of recommendation methods have been developed, construcingmore intelligent and more robust recommender systems is still faced with many theoreticaland technical problems which have not been well resolved. In this background, this articlestudies the relative recommendation methods and systems, and proposes a novelrecommendation algorithm. The main work and innovation of this article are summarized asfollows:(1) This article presents a general review of current research of recommendation systemsand algorithms, including basic principles, key technologies, advantages and shortcoming,especially the main similarity calculation and rating estimation methods. Simultaneously,summarizes the major evaluation metrics, and point out the key problems of recommendationsystems.(2) Two new similarity computing methods have been proposed in this article.Collaborative filtering is the most commonly used as a recommendation algorithm, and thesimilarity calculation method is its core. Existing similarity algorithms fails to give fullconsideration to the meaning of the rating, and ignores the two important aspects of users’attitude with liked or disliked and the degree. To solve this problem, according to theconsistent attitude of the users on the items, proposes the similarity calculation method ACS.In addition, taking full account the two meanings of the rating, proposes the LRWS methodbased on the finite-step random walk. The experimental results on the Epinions dataset showthat ACS similarity superior to the other similarity in the comprehensive consideration of themean absolute error (MAE), root mean square error (RMSE).The experimental results on theMovieLens dataset show that LRWS similarity superior to the other similarity under the MAEand RMSE.(3) A new algorithm FTRA has been proposed, which infuses users’ trust network andrating data. The sparse problem of rating data will significantly reduces the accuracy of collaborative filtering recommendation. In addition to the users’ ratings data on the Internet,other data sources which can be used in the process of recommende, and one of the morecommon is trust network data which describes the mutual relationship between users. Tosolve this problem, this article will the data of trust network as an important supplement onthe rating data, and bases on graph theory concepts or methods, the similarity method in thethird chapter, and the Katz method which is used to calculate the similarity of link, proposesthe FTRA algorithm which organic infuses this two data, and then better to solve the sparseproblem of the rating data faced by collaborative filtering. The experimental results on theEpinions dataset show that the FTRA algorithm is superior to or significantly better than thecomparison algorithms, which include the algorithms that only based on the rating data or thetrust relationship, and the other algorithms infusing the two data sources.