Users´ brands preference based on SVD++ in recommender systems

Recommender systems provide users with personalized suggestions about products or services. General task of recommender systems is to improve recommendation accuracy, but this paper mostly focuses on improving the degree of surprise, using SVD++ (singular value decomposition) model. First, logistic regression method is used to process raw data including different sorts of user actions on brands, such as click, shopping cart and buy, so that user-brand ratings are obtained. Then SVD++ model is used to analyze the processed data. A better RMSE(root mean square error) is achieved through adjusting parameters, so the system recommend new brands which users have no actions before to improve users´ the degree of surprise. Model presented here is applied to analyze Tmall data, and the result proves its efficiency.