Temporal Dynamic Characteristics Research And Distributed Implementation Of Recommendation Algorithms

In recent years, with the rapid development of information industry, the amount of data being produced are exploding, that leads us suffering from "information overload" problems. The purpose of advent of recommended system is to solve the problem of information overload. However, the traditional recommendation algorithms does not take dynamic characteristics into consideration, that makes low performance. At the same time, the real-time performance of the recommendation algorithm is challenged by the large amount of data. The main tasks of this paper are list below:Research on temporal effect and dynamic characteristics of recommendation system such as user’s interests change, user’s ratings change, seasonal effect, and summarized the research progress of dynamic characteristics. And research on modeling user interests change and product popularity into matrix factorization. Experiments on Netflix and Movielens datasets shows, the accuracy of the algorithm called "TemporalBMF" is improved significantly.Research on parallel alternating least squares (ALS), stochastic gradient descent (SGD) and distributed stochastic gradient descent (DSGD), which optimize the matrix factorization algorithm. Implement matrix factorization with ALS, SGD, DSGD. The experimental results show the feasibility and validity of the parallel algorithm...