Research On E-commerce Recommendation Algorithm Based On SVM Model

With the popularity of the Internet and the development of e-commerce, e-commerce system provides users with more and more choice, but at the same time the structure has become more complex. Users often get lost in the space with a large number of goods, and unable to find the merchandise they need. So e-commerce recommendation system came into being, it can directly interact with the user, and provide users with recommended commodities to help users to find what they really need to buy. The existing recommendation algorithms in e-commerce recommendation system are discussed in this paper and two new model-based recommendation algorithms using support vector machine aimed at different user types are put forward. They can appropriately solve the sparsity problem and scalability issues in the original recommendation system.Firstly, due to the superior performance of the support vector machine on classification problems, the linear support vector machine classification method is introduced to recommendation algorithm in this paper, and the experiments are executed from both the points of user-based and item-based. Experiment results show that the support vector machine classification algorithm has obvious advantages on relatively sparse data. This method can be used for many times to model once, it has good scalability. In addition, this method can be used to recommend for the old customers with more shopping records because a lot of rating data must be used.Secondly, because of the importance of hot commodity, a recommendation algorithm aimed at hot commodity with support vector machine regression based on user characteristics is proposed in this paper. The algorithm constructs customer group feature model of the hot commodity using their personal attributes and behavioral characteristics, then the relationship between the model and the actual score is mined by utilizing the support vector machine regression algorithm. Compared with the conventional collaborative filtering algorithm, this algorithm can improves the accuracy of the recommendation with smaller mean absolute error value. This algorithm is more appropriate to recommend for the new customers with fewer shopping records and it can to some extent solve the data sparsity problem and cold start problem for new users.Finally, design of the recommendation system model is proposed combining with the two algorithms proposed. The framework of the entire e-commerce recommendation system is built, in addition, the modules are analysis and the database storage is given. This prototype cin better meet the needs of different users.