Improve E-Commerce Recommendation by Classification Tree and Fuzzy Sets

Abstract

In order to enhance the performance of E-Commerce recommendation, a hybrid filtering approach based on the taxonomy of E-Commerce platform is put forward. The classification tree of products is used to find the users with similar shopping intention. The sparsity of user ratings, major problem for collaborative filtering, is overcome. A two-granularity user profile is built to reflect the customer's shopping interests. User profile is firstly described as a set of leaf nodes of the classification tree. Then, each category of the user profile is refined by the theory of fuzzy set. Fuzzy sets make user profile and item representation more accurate. At the same time, tags instead of key words extracted from item content, are used for the building of user profiles and representation of items. It overcomes the analysis difficulty and large calculation problems for content-based filtering.