Prediction Of Purchase Intention Of High Potential Users Based On Data Mining

In recent years,"big data" has attracted the attention of many scholars and enterprise managers.The emergence of massive data has ushered in a major era of transformation.In the era of big data,people are required to use the thinking of big data to process and analyze increasingly large data resources in but not limited in the area of healthcare,education,business.Domestic and foreign scholars have proposed research methods and application examples for big data.Companies are also turning their attention and resources to developing products based on big data.The Internet e-commerce industry is one of the most representative areas.Jd.com has officially entered the e-commerce in January 2004.In recent years,it has developed particularly rapidly,and now it has become the largest self-operated comprehensive online retailer in China.With the increase of the popularity of the platform and the increase of scale,the number of users of JD has increased dramatically,and now JD has accumulated a large number of user data.So,how to find valuable information from huge amounts of data,to predict the future purchase intention of consumers,is the key to user big data to solve the problem of precision marketing for not only JD but also the whole e-commerce.This study is based on the actual behavioral data of more than two billion users in JD.In the early stage,this paper focused on feature engineering.We Extract data from the Hadoop cluster with the help of Hive SQL,and then the characteristics of 646 dimension were determined,which laid a good foundation for model prediction.Secondly,with the help of Apache Spark to complete the processing and calculation of data.And based on the logistic regression algorithm,a short-term intention model is constructed to predict the purchase intention of users and output the prediction probability of users purchasing a certain category.In the single experiment stage,using JDK8,Spark v2.1.1,Hadoop v2.6 and PyDev plug-in to develop python programs on Eclipse Neon3.Finally,it gives the high-potential users the product category coupons they have predicted,helps them to buy the products they need.After the continuous optimization of features and models,it has been verified that the predict model we built has obtained great results.The exact rate and recall rate are around 0.28.After the model was launched,the platform added more than two million orders a day.This research shows that e-commerce platform is based on big data to predict the feasibility and practicability of users' purchase intention.In the future,with the introduction of more complex features and more complex prediction algorithms such as cyclic neural network and deep neural network,the prediction accuracy of the model will be further improved.