Abstract:

This paper presents a novel knowledge tree based recommendation algorithm，which is different from traditional approaches.At first we construct a word based knowledge tree using resources from the Internet，and then we employ EM algorithm to the tree.We automatically update the knowledge tree using unlabeled documents，making classification of both the terms and document optimized at the same time.Since the algorithm does not rely on any human annotations，it is an un-supervised machine learning algorithm.In real applications，user interactions in the mobile Internet are mapped to certain categories of the knowledge tree，and similar products are recommended to users according categories in the tree.Experiments in the mobile e-commerce network community project indicate the high efficiency of our algorithm，and the performance of user satisfaction achieves about 83%.

Key words: mobile e-commerce, personalized recommendation, knowledge tree, EM algorithm