Research On Revisit Prediction Based On User Behavior Data

With the rapid development of e-commerce and the continuous improvement of mobile terminal technology,online shopping as a new way of consumption is becoming more and more popular.Tens of thousands of online browsing behaviors happen every day.People can browse online whenever and wherever possible,or they click,or add to favorite,or add to cart,or purchase,so the field of electronic commerce gradually accumulated a large amount of log data of user behavior,effectively mining the data and get valuable information now is a hot research topic.Some users will revisit e-commerce platforms or merchants on e-commerce platforms,and predict user's revisit behavior based on user behavior log data,which is of great value.Precise and personalized re-visit forecasting,contributes to e-commerce platform taking the whole situation into account and plan accordingly,reasonable arrangement of resources,intelligent management of the shops,providing better service to users;helps merchants to conduct accurate marketing and attract consumers through the distribution of shopping vouchers to increase revenue,reduce costs,and increase return on investment;helps customers to meet their own interests and receive personalized recommendation,save the time and money cost etc.Based on the customer basic information data and user behavior log data of the e-commerce platform,this paper conducts an in-depth study of user revisit prediction problems based on the deficiencies of the existing revisit prediction work,and proposes two revisiting prediction models.The work and contributions of this article include:1.We propose a platform-based revisit prediction model based on Hidden Markov Model.The model uses a hidden Markov model to study the user's revisit behavior to the e-commerce platform.Specifically,first,the observation sequence data is known,and the parameters of the model are learned using the Baum-Welch learning algorithm,which contains the hidden data state sequence;Then based on the model that has been learned in the previous step,and then known observation sequence data,we use the forward algorithm to calculate the possible behavior of a user at time t+1.We validate the proposed model based on real e-commerce data.The experimental results show that the model predicts that the user's revisit to the e-commerce platform is effective.2.We propose an accurate revisit prediction model for users based on ensemble learning.This article first introduced the ensemble learning-based revisit prediction algorithm.Specifically,the method first preprocesses the log data of the original e-commerce platform user behavior;then,based on the preprocessed data,the revisit prediction feature engineering is established from three aspects:user,merchant,and user-merchant interaction;Then we use the ensemble learning algorithm Stacking to perform user-to-merchant re-visit prediction based on the proposed re-visiting related features.Some basic statistical learning methods are used in ensemble learning methods,including decision trees,random forests,logistic regression,and neural networks.At the same time,this method also solves the problem of sample class imbalance and the optimal threshold selection problem of voting mechanism.Finally,based on this method,the user's precise re-visit prediction experiment for the merchant is performed.The experimental results show that our method is more accurate than the baseline method.