Research On Electronic Commerce Recommendation System Based On Collaborative Filtering

This paper researches on the personalized recommendation systems current widely used in electronic commerce and introduces general situation of the research on personalized recommendation systems at home and abroad. The paper focuses on user-based collaborative filtering recommendation system which is the mostly used and analyzes its working principle. To aim at the existing problems such as the sparsity of user evalution, the low storage space utilization, the shortcoming of self-adaptive, this paper raises a solution as an improved collaborative filtering recommendation system.This paper adopts an improved cross list storage structure to store data elements in user-resource evaluation matrix, which not only can support the dynamic change of the matrix, but also can compress system storage space in the greatest degree. At the same time, this paper brings the theory of Semantic Web and Ontology into collaborative filtering recommendation system, predicting the score of rescource that user hasn't given an explicit score with the help of semantic relations between resources, which solves the problem of low recommendation precision brought about by sparsity of evaluation to a certain extent. In addition, the improved system will record the results and refresh the user-resource evaluation matrix synchronously, which can take full advantage of previous results to increase recommendation precision gradually and then improves the system adaptability effectively. In this paper, statistical precision of measurement method is used to compare the traditional user-based collaborative filtering recommendation system and the improved collaborative filtering recommendation system in order to prove the improvement at system recommendation accuracy.