Abstract: In the field of personalized recommendation，current algorithms have the disadvantages of lower precision and deficiency on recommendation.This paper presents an algorithm based on the features of product，the customer’s purchased logs and real-time browsing action.Firstly the content of on-line browsing is collected to deduce the purchase preference of current customer，then contrasts the purchased logs and the database of product features and analyzes them，by means of which，the preference degree of the product features and corresponding commendation reference groups can be obtained，therefore according to the similarity matrix of features entity，reference groups are recommended.Finally，integrated with the purchase preference and the former results，products are recommended to the customer.