Distributed design and implementation of SVD++ algorithm for E-commerce personalized recommender system

Abstract

Recommender systems can facilitate people to get effective information from the massive data, and it is the hot research currently in data mining. SVD++ is a kind of effective single model recommendation algorithm, which is based on the matrix decomposition combined with the neighborhood model. On the Spark, using the Stochastic Gradient Descent, this paper realized the distributed SVD++ algorithm through the Scala, deployed and applied the algorithm into an actual recommendation product for testing. The testing results represent that the distributed SVD++ algorithm succeeded in solving problems of terabytes of data processing in the e-commerce recommendation and the sparse data of user-item matrix, enhancing the quality in personalized commodity recommendation.