Design And Implementation Of E-commerce Website Behavior Analysis System

The user's click and browsing behaviors are recorded in the Web server log.Mining the website log,analyzing the user's access behavior,and optimizing the E-commerce website are research hotspots in recent years.Generally,the studies of user preferences and conversion rate are two major aspects of user behavior analysis.There is a large number of studies on user preferences which have been applied on personalization feature of website.However,the industry has made little research on the optimization of conversion rate and website structure in E-commerce websites.This two issues are addressed in this study and the scalability problem of massive log processing for large-scale E-commerce website is also solved.The system presented in this study includes a module of click path construction,a module of conversion rate analysis and a module of information structure assessment.The specific research content of these modules includes:(1)The user's click path is the basis of user behavior analysis of E-commerce website.A method of constructing user click path from unstructured original log is proposed which uses Map and Reduce programming model to solve the scalability of massive Web log analysis.(2)The funnel model has the disadvantage in optimizing the conversion rate.The reason why the conversion rate is low cannot be analyzed according to the user click path.A conversion rate analysis method that combines a funnel model with frequent pattern mining is proposed.The minimum single-step conversion rate found by the funnel model is used as an input of frequency threshold to frequent pattern mining.Therefore the possibility of mining the related frequent paths for websites can be maximized.These information can help enterprises to identify key points that should be optimized combined with the information of the user's click path.(3)The deep neural network model is introduced to analyze the rationality of the structure of website.First,user click paths are vectorized by word2 vec,then the user's click distribution can be predicted with the LSTM.This result can be analyzed combined with the expected result of the website to optimize the structure of the website.