An Improved Recommendation Model Based on Matrix Factorization
Song-Wei Wei, Lijuan Zhou · 2021
Matrix factorization techniques has become a dominant methodology within collaborative filtering recommenders. It has proven to be efficient in recommender systems when predicting user preferences from known user-item ratings. One of the matrix factorization benefits is very good at discovering the potential interests of the users. In a movie recommendation system, a common phenomenon is often to recommend you some movies that you have already seen. Unlike music recommendations, users may listen to the same music repeatedly, but they hardly watch the same movie repeatedly. Recommending a lot of movies to a user that user has already seen is a serious problem. In this paper, we improve the Latent Factor Model to reduce this situation. At the same time, high-quality films increase the chances of being recommended. Finally, we introduce how to build the movie recommendation system with Apache Spark. The model we proposed reduced the problems described above by reducing the prediction score of popular films and increasing the prediction score of high-quality films which can improve user experience.