Research On Personalized E-commerce Recommendation System Based Mixed Algorithm

With the development of Internet and E-commerce, consumers may be confronted with the serious problem of information overloading when they choose commodities. Therefore, many E-commerce web sites research and develop recommendation system to provide consumers individual recommendation services. The recommendation system works like salesman who gives consumers advices and helps them to find what they need. With the wide using of the systems, many problems and challenges come out. For examples, the conflict of recommendation quality and real time, sparsity of data, cold start and etc. This paper researches on the recommendation systems and recommendation method which is the heart of the systems to solve these problems.Firstly, the paper researches on the recommendation systems. According to the problems of recommendation quality and real time requirement, the system must be built more reasonably. It is composed of two parts, the on-line real time recommendation part and the model processing part. The first one presents the recommendation list to the users when they browse the E-commerce web sites. The model processing part collects data and uses different models to have the outputs of models which are the bases of the on-line part. Especially, the system has the methods to recommend the new users and new commodities. The structure of the system can help to solve the cold start problems and improves the recommend quality.Secondly, the paper researches on the recommendation methods. Existing methods include content based recommendation, collaborative filter, mixed method, data mining and etc. Among the above, collaborative filter is used more frequently and successful. Therefore, the paper uses the Fuzzy ART algorithm based on collaborative filter and data mining. The experimental results show the algorithm has better recommendation quality than the old method.Thirdly, according to the sparsity of data and cold start problem, the paper advances the Fuzzy ART algorithm with the content based recommendation method. The new method uses the information of commodity characters and interest of users to process the interested model. The model is classified by Fuzzy ART and then it predicts the result of recommendation. Compare with Fuzzy ART, the new method helps to solve the sparsity of data. Otherwise, it improves the recommendation of new commodities.Finally, consider the influence of user information to the recommendation quality, the mixed method is advanced further. The basic information of users like age, occupation and gender are related to the behaviors of purchase. The mixed method uses the information to adjust the result of classify. The experimental show that the mixed method is the best of the other ones. Meanwhile, the method can solve the cold start problem about new users who have not any interesting data. The method only uses the information of users to find their similar neighbors in order to give them advices.