Matrix Factorization Recommender System Algorithm Combing Personal Compact And Trust Propagation

In recent years, with the rapid development of Internet technology, people enjoy the rich and colorful data, but also suffered “information overload” issue. The recommender system can solve this problem to a certain extent, the recommendation system based on the user and the relevant information of the project, and then recommend to the user may be interested in the project. Collaborative filtering is the most widely used recommendation technology, based on neighbor method and the method based on the latent factor model is collaborative filtering technology in two main areas. Among them, matrix decomposition is based on the latent factor model used in the successful realization of, the research work mainly focus on the matrix decomposition was carried out. The research results are as follows.Firstly, a trust transfer model is established for the purpose of alleviating the sparsity problem of user’s trust data. The idea of using the trust transfer, filling the user trust data, enrich the trust data, and then use after filling the user trust data to predict the missing data of user rating data, in order to alleviate the sparsity of user rating matrix. A trust transfer matrix decomposition recommendation algorithm is proposed, which is compared with the traditional matrix factorization algorithm, and the algorithm has better prediction effect.Secondly, an individual influence calculation model is established, which aims at improving the accuracy of the system’s recommendation. Considered inter individual differences, with thought of star effect and factors into social groups of influential individuals, using the PageRank algorithm, use and the degrees of the concept, in order to calculate the user individual influence, proposed a fusion individual influence matrix decomposition recommendation algorithm. The experimental results show that the accuracy of the system is improved.Thirdly, a matrix decomposition algorithm is proposed to preserve the original data structure. Using the projection structure of Non-Negative matrix factorization algorithm, combining the trust propagation model and the personal compact calculation model and user rating prediction model, we proposed a fusion of influential individuals projection of the structure of Non-Negative matrix factorization recommendation algorithm to improve the stability of the recommendation system. Experiments show that the algorithm compared this algorithm not only reduces the MAE and RMSE, but also improves the prediction accuracy of the system.At last, the design of the distributed computing algorithm. Using the open source Hadoop large data processing framework, using the simple and efficient MapReduce programming model, the design of the algorithm TP-SPNMF distributed computing.