Research Of O2O E-commerce Recommendation Model On Gradient Boosting Regression Trees

With the popularization of mobile smart phones, mobile 020 e-commerce is ushering in a broad market. Therefore, the mobile end consumption is more popular and higher demands are made on the recommendation system. Based on the recommended model in current 020 scenario, a more specific and timely personalized recommendation system which can update new recommended content with the fast changes of the move location is requested.Based on the research on 020 mobile e-commerce mode and Alibaba’s real transaction data of livelihood service goods for a month, this paper has found out that customers will make periodicity consumption towards some kind of service goods because of their own life style;secondly, the Matthew effect will occur in the hot consumption scenario and mobile subscribers’ average propensity to consume will increase greatly. Lastly, the propensity to consume in current position scenario has affected the percent conversion of the goods in current scenario.Based on the above three assumptions, the preferential model is built for the subscribers based on their location information, and a recommended model of 020 e-commerce is put forward based on the gradient regression algorithm, in which current time parameter and location parameter are introduced. In the meanwhile, the users’behavior log in 020 e-ecommerce is mined and several traits which can distinguish the customers’ purchasing behavior toward commodity and service are extracted. Then, these traits are integrated into the gradient regression algorithm to predict the customers’ purchasing behavior.Experimental results show that the modified 020 e-commerce recommended model which is based on the gradient regression algorithm is evidently superior to the traditional recommended algorithm in instantaneity and veracity.