Cross-domain Recommendation Algorithms And Applications Via Tensor Decomposition

In recent years,the Internet,especially the mobile Internet develops rapidly.The major Internet companies derive a large amount of data about users' lives and behavior patterns.The way to use these data,dig out the useful information from a variety of heterogeneous data and recommend the user related products becomes a problem.From the current perspective,academia and industry in the field have a more profound study about singledomain personalized recommendations.The mainstream approach academia and industry solving the recommender system problem is based on single-domain recommendation.It predicts users' behavior and preferences by learning historical data of the designated domain.However,single-domain recommendation does not work well with sparse data problem.For this reason,cross-domain recommendation,which considers combining multi-domain data to help predict target domain has become a hot topic.This paper first describes the lack of single-domain recommendation and advantage of cross-domain recommendation,emphasizes the significance of cross-domain recommendation,and describes the issues as well as challenges of cross-domain recommendation,presents and analyses of the current major cross-domain recommendation algorithm.With this knowledge,this paper focuses on cross-domain recommendation via tensor decomposition.This paper discusses the theoretical basis of tensor decomposition,seeking its cross-domain recommendation scenarios based on its characteristics.The main work of this paper is reflected in the following two aspects:1.Cross-domain Flight Recommender System via Tensor DecompositionThis article describes the background of flight recommendation,demand a clear recommendation flight problems and the difference with traditional recommendation.This article then introduces a simple idea based on cross-domain triadic factorization.The algorithm is improved by adding history weighting strategy to solve the problem of crossdomain personalized flight recommendation.In addition,this paper presents the idea of how to parallelize this algorithm.2.Cross-domain New Movie Recommender System via Tensor DecompositionThis article describes the background of new movie recommendation with a clear need for new movie recommendation issues.This article then describes how to use the idea of single-domain recommendation to solve new movie recommendation and its deficiencies.Then we solve the new movie recommendation problem based on cross-domain triadic factorization algorithm with implicit feedback.In addition,this paper proposes a method of optimizing the storage and computation of this algorithm based on new movie recommendation features.