Research On Spark-based Personalized Recommendation System

With the continuous development of big data,it becomes very meaningful to know how to extract data that is valuable to users from big data with the quick,efficient,and accurate way.The recommendation system is one of the effective methods to solve such problems.For the recommendation system,there are problems of low accuracy and low coverage.This paper optimizes the user-based and item-based personalized recommendation algorithm,and implement the optimized algorithm on the Spark platform.The experimental results verify the performance of the optimized algorithm,the research work and content of the paper is as follows:(1)In this paper,the user-based collaborative filtering algorithm has low recommendation accuracy,and the similarity between traditional computing users is improved.The main feature is to introduce the attribute feature vector of the user's favorite items.The similarity between the user's collaborative filtering similarity and the attribute eigenvector of the user's favorite item is taken as the similarity between the users,and the pre-optimization's algorithm and the optimized algorithm are implemented on the Spark platform,and the corresponding experiments were carried out to analyze.It proves that the optimized user-based collaborative filtering recommendation algorithm improves the accuracy of recommendation.(2)In this paper,the item-based collaborative filtering algorithm has the problem of low recommended coverage,and mainly improve the similarity between items's computing by introducing parameters into the similarity formula between items,and the pre-optimization's algorithm and the optimized algorithm were implemented on the Spark platform.Finally,the corresponding experiments were carried out to analyze,and the optimized item-based collaborative filtering was proved.The recommendation algorithm improves the recommended coverage.