Design And Implementation Of Electronic Business Recommender System Based On User Behavior

The emergence and popularization of e-commerce system has brought great convenience to the users and satisfied the demand of consumers in the information age. However, it brought users the explosion of information by the rapid development of information technology and the accumulation of large amounts of data. Facing the problem of information overload, users can not find their interested goods. The recommendation system is an efficient method to solve the information overload problem and provide the personalized commodity information for the user. Based on the user’s individual needs, social relations, and demographic attributes, recommend system could recommend products to users, and show users convincing reason for the recommendation.Therefore, the recommendation system allows the e-commerce system to effectively improve user retention, promote user activity and increase sales.Based on the large amount of user behavior data and low value density,this thesis designs and implements an electronic business recommender system which is suitable for dealing with user behavior data. The main contributions of this thesis are as follows: (1) A hierarchical recommendation system architecture is proposed, which is characterized by high reusability and low coupling. (2) In the traditional recommender system the recommendation algorithms are mainly collaborative filtering algorithm, popular recommend algorithm, etc. In this thesis, the recommendation algorithm of purchase probability prediction is proposed. There are logical regression and factorization machine as kernel model of the recommendation algorithm available for selection. (3) In the recommendation algorithm of purchase probability estimation, feature engineering has always been a key part. In this thesis, two schemes of artificial feature combination and machine learning automatic feature combination are studied and compared.