Research Of Learning Optimal Ranking With Tensor Factorization For Recommendation System

The recommendation system is a kind of information filtering technology which widelyused on the internet. In order to provide better service for users, many famous websites suchas Amazon and Facebook have integrated recommendation system. Recommendation systemrecommends products and services that a user may be interested in.The traditionalcontent-based recommendation system used the features of users and items to recommend.Collaborative filtering recommendation system predict what users will like based on theirsimilarity to other users.Folksonomy allows users to use tags to make personalizedclassification and mark the products. How to use user-item-tag ternary relationship andimprove the accuracy of recommendation system has been paid more attention.Based on the ternary relationship, we proposed a recommendation algorithm named"Learning Optimal Ranking with Tensor Factorization for Recommendation SystemAlgorithm"(short for “LORTF”). LORTF algorithm contains two main steps. Step one, a usercluster is built to reduce the size of data set by defining core user. Step two, calculating theweights of ternary relationship, building a three dimensions tensor model, using the TuckerTensor Decomposition to factorize tensor with optimization method. Finally the new optimalscores are obtained according to the approximate tensor and a Top-N list for recommendationis generated.We do experiments on five real world data sets, which are MovieLens, Flickr, Delicious,Last.fm and Bibsonomy.The experimental results demonstrate that LORTF algorithmoutperforms other state-of-the-art recommendation algorithms.