Research On Collaborative Filtering Recommendation Algorithm Based On Matrix Decomposition

The rapidly developing Internet in today’s society is an important way for people to obtain information, while information overload has become a problem that people have to face. Recommender System is an effective tool to solve the information overload, which is a hot issue in today’s academic and industrial research. Collaborative filtering recommendation algorithm is currently the most widely used algorithms, matrix factorization based methods in recent years has received more and more attention.This paper focuses on the matrix factorization based collaborative filtering algorithm in deep study and discussion.1. It comprehensively introduce the background and significance of research recommender systems and the research status at home and abroad. Then introduce the common recommendation algorithms in details, including Content-Based recommendation, Collaborative Filtering recommendation, Map-Based recommendation, Association-Rules based recommendation and Hybrid recommendation, and then outlined the challenges Recommender System faces.2. Details of the state-of-art recommendation algorithm based on matrix decomposition are described, including basic singular value decomposition, Regularized singular value decomposition, Biased singular value decomposition, singular value decomposition adding neighborhood information, singular value decomposition adding time information and non-negative matrix factorization methods.3. For discrete rating datasets, an improved binomial matrix decomposition model is proposed, which is under the assumption that the data subject to binomial distribution.The loss function is learned from Maximum A Posteriori estimation, and solved by using the stochastic gradient descent algorithm. And the rounding issues of ratings were discussed, a new rounding algorithm based on global optimum was proposed.4. This paper introduces the concept of trust in sociology, gives a method to calculate global, which is applied to matrix factorization model in this paper, and gives the solution process of the algorithm.5. Evaluation of the recommendation system is briefly introduced. We evaluate the algorithms proposed in this paper on Baidu movie recommendation and MovieLens datasets, and the results are analyzed.