The Research Of Hybrid Recommendation Algorithm Based On Clustering And Session Scenarios

With the popularity of Internet technology, e-commerce has become an indispensablepart of our life. At the same time，information overload, resulting from the excessivecommodity information, make users can not access information of interest effectively, andreduce the efficiency of the use of information. Search engines, an information retrievalsystem, is an important information to help users filter means, but the search engines can’tmeet the user’s individual requirements by providing information on the way to take theinitiative. Personalized recommendation system is not only able to meet the user’sindividual needs, but also the most effective tool to solve the information overload.Content-based recommendation and collaborative filtering recommendation are twocommon personalized recommendation algorithms. Content-based recommendations buildthe user’s interest model based on merchandise users has rated, and then recommendeditems which meet the model to the user. Collaborative filtering recommendation calculatesthe predicted rating of product through the neighbor’s score, and recommends the highestscore of products to the user. Both recommendation recommended product to the userwithout considering the real time and diversity of user requirements.This paper presents a new hybrid recommendation algorithm applied to e-commerce.The algorithm first complete the user clustering process, and then convert to the user’sbehavior on the product pages for product ratings, compute the similarity of the clustercategory of users, and finally get prediction score of ungraded goods. When the user usingthe recommended system, the algorithm get the current user’s purchase requirements，through analyzing products which has been browse in the current session. Evaluating theproduct collaborative filtering ratings based on the purchasing requirements，the algorithmget a hybrid predicted score. The experimental results show that the proposed algorithmcan improve the efficiency of recommendation significantly and get a more accuraterecommendation result.