User Profile Research Based On User Consumption Behavior

The value of massive data in the modern business era has become more and more obvious and preeminent and enterprises engage in digging out user's demand from the massive data,which make user profiles recently a basic application for enterprises to studying user.Different business scenarios require different user portraits.User profiles based on consumer behavior enable enterprises to provide accurate service,personalized marketing and featured recommendations to users.This paper aims at building user profiles based on of consumer behavior.In this paper,we analyze the consuming data to determine the labels of the profiles.On top of that we propose a user profiles model,including a user's consumption ability predicting model and the user consumer behavior analysis model,both based on stacking.This model applies feature engineering,including feature extraction and feature selection,to consuming data.The stacking-based user's consumption ability predicting model is comprised of two layers.This paper presents an integrating tragedy of selection of the individual learning units,blending the accuracy and difference of the individual classifier to combine the modules from the first layer.The user consumption behavior analysis model combines user clustering model and user consuming behavior association rules mining model,which cluster the user according to their consumption abilities through Gaussian mixture model and dig the association rules from the user cluster in a way combining single-dimensional and multi-dimensional association rules.The experimental results show that the stacking based user consumption behavior model outperforms the single classification models and traditional integration models.Selecting the appropriate number of Gaussian distributions based on the Bayesian criteria makes the user clustering model based on Gaussian mixture model be not only better than the density-based DBSCAN model and partition-based K-MEANS model,but also more suitable for dividing the user's scenario.Consumption industry association rules mining,combining single-dimension and multi-dimension association rules,effectively get the association rules of consumption industry,to analyze the user's preferences.The research of this paper shows that it is of practical and pragmatic value to build user profiles based on user consumption data.The feature processing of user consumption data in specific scenarios and the analysis of user consumption behavior is of significant importance to future works.