Research On E-commerce Personalized Recommendation Algorithm Based On Collaborative Filtering

With the rapid development and popularization of the Internet, modern e-commerce has been widely applied, more and more product information filled in the business website, people are faced with so many goods feel helpless, it is difficult for them to find the goods they really need in a short period of time, which is called "information overload" phenomenon. In order to help customers quickly find the goods they need, e-commerce personalized recommendation technology has come into being.At present, collaborative filtering is one of the wider application and recommended better technologies for personalized recommendation systems. This thesis is based on the collaborative filtering recommendation techniques, to improve the recommendation quality as a starting point. Against the traditional User-based collaborative filtering’s some shortcomings, the thesis gives an improved User-based collaborative filtering algorithm. Compared to the traditional User-based collaborative filtering algorithm, the improved algorithm has some advantages. First, the original user-item rating matrix is too sparse to affect recommended quality. Improved User-based collaborative filtering algorithm fills the original user-item rating matrix. This approach not only retains more useful information, but also improves user-item rating matrix destiny. Second, in the calculation of nearest neighbors, the improved algorithm takes into account the nature of user interest. Third, the improved algorithm adds the time element, gives the later rated items a higher weight, so it can reflect the trend of user interest. By this way, the improved algorithm can get more similarity neighbors to the recent interest of the active user. Fourth, the improved algorithm classifies the items, finds neighbors in the sub-class including the items the active user liked, in every sub-item rating matrix produces sub-candidate recommendation collection, the last combines all sub-candidate recommendation collection as the final recommended list. By this way, the improve algorithm can find more "true" neighbors to generate more accurate recommendation to the active user.In this thesis, some experiments are designed to validate the effect of the improved algorithm. These experiments use the Movielens set、EachMovie set as the experiment data and the MAE to analyse the recommendation quality. At the end, the results show that the improved algorithm can improve the recommended quality. It has better prediction accuracy.