Study On Electronic Commerce Personalized Recommendation Model

With the application and development of Electronic Commerce, integrated researches of artificial intelligence, web mining and commercial model have become a forward problem. Electronic Commerce websites provide more and more options for customers, at the same time, increase the difficulty that customer, who face with plenty of information, find out the product information coincide with their needs accurately and quickly. Personalization recommendation technology analyses customers'related information, and offers products and service to customers which coincide with their preferences actively and efficiently. On the one hand, the customer personalized demand was satisfied much more better. on the other hand, it is favorable to establish steady customers crowd, improve service quality.This dissertation analyses research accomplishment and actual application environment, studies the personalized recommendation model which was based on customer purchase behaviors and preferences, and filters customer samples with sampling technology in order to improve the accuracy and efficiency of recommendation. Research accomplishment mainly includes:①This dissertation offers a dynamic method to mine customers behaviors. Because most of traditional technology predict customer preferences based on the static data, but customer preference is changing with time. So this dissertation sorts the purchase of customer behavior as purchase behavior sequence, according to the time order, and then extracts the association rule, and predicts target customer's current or future preferences, which has raised the accuracy of the forecast.②This dissertation offers sampling for pre-processing sample data. This dissertation takes look ahead selective sampling algorithm and collaborative filtering algorithm based on the combination of project and customer in use. Computer, through defining sample label utility, chooses the customer sample with the maximum utility for labeling as the recommendation basis. And it solves sparsity and extensibility of traditional collaborative filtering, which provides a way for reducing cost and raising quality of recommendation.③This dissertation puts collaborative filtering and collaborative filtering basde on project and customer into constructing experiment models, to evaluate the validity of the personalized recommendation model. It puts testing ensemble data of Movielens in use, analyses mean absolute error and standard variance of mean absolute error, and predict time as experiment quotas which indicators that the new model can provide more accurate and swif recommendation result.The purpose of the article is to solve the contradiction between the mass product information and the individualized requirements of users, carry out a method to recommend the exact products to different users according to their interests, finally, accelerate the development of personalized and intellectualized information services in E-commerce of China.