Content-based And Behavior-based Filtering Complement Recommendation In E-commerce

With the rapid development of the Internet industry,people have entered the era of information overload.Under the background of the massive Internet information,it is difficult for consumer to find interesting information from a large amount of information.Major Internet companies also found it is hard to display information which users are interested in to the users.Personalized recommendation system can recommend information which users are interested in to the user.It can explore the long tail items to the users.So it is undoubtedly an effective solution to the problem of information overload.Nowadays,in E-commerce recommendation field,item-based collaborative filtering is widely used in the large business website.such as Taobao,Amazon etc.However,because computing the similarity relation among items depends on user behavior data,recommended result is poor in novelty and exploring ability.In addition to substitute relationship between the items,there is another kind of relationship called complement between items.It can effectively improve the ability of discovering of the recommendation system.In this paper,the proposed Complement recommendation algorithm consists of two parts,one is collaborative bought filtering approach.The other is modeling based on the product category information and text information.We use topic model and label propagation model.Eventually we combine the two parts.Experimental results show that the proposed recommendation algorithm based on content and behavior can not only greatly enhance the accuracy of the collocation,but also effectively enhance the diversity,novelty and coverage of recommendation system.