Consumer Behavior Analysis And Application Reasearch Based On Deep Learning

In recent years,with the rapid development of E-commerce,online shopping has gradually become the main consumption method,while consumers enjoy convenience,problems that long browsing time and low purchase conversion rate continue to emerge.Therefore,E-commerce companies use machine learning models to analyze consumer behavior and build recommended systems to accurately recommend products.Consumer behavior analysis is significant to improving the efficiency of the recommended system and improving consumer satisfaction.Considering the high request for recommending accuracy,customers' satisfaction and working efficiency of the recommended system,this article refine the research details including data processing,feature engineering,classification prediction model and application prospects.Take the behavior datas of the consumers as the experimental subjects,building traditional forecasting models and deep learning models frameworks respectively.It can prove the deep learning models' much bigger protencial for learning by comparative analysis on evaluation methods.To sum up,the main work of this article is as follows:(1)Research on behavior data processing and feature engineering.Starting with the data processing flow,it focuses on research on extracting features artificially,feature preprocessing,feature selection and random sampling techniques for unbalanced data.(2)Comparative study on traditional models.This article constructs Logistic regression model,random forest model,neural network model.The results of the three models show that the random forest model has stronger predictive ability,however,it may generate a tree with small difference during the operation,affecting the correct dicision making.In view of that logistic regression model does not apply to high randomness,nonlinear data prediction problems,while shallow neural network model is easy to fall into local minimum during training,the traditional models can only learn the shallow characteristics of datas,which will affect the prediction effect to some extent.(3)Research on deep learning models.The purposes of this article are to learn the deeper features of the datas and to improve the accuracy of the prediction.This article constructs deep learning prediction models,the experiment confirm that DNN model can improve 10% efficienciness compared to traditional models,but DNN model runs slowly and it's training requires high costs.Therefore,in order to overcome the shortcomings of DNN model,combining the characteristic of unbalanced distribution of data labels,the article proposes rDNN model.The training time of rDNN model is greatly shortened to 434.361 seconds,improving the computational efficiency of DNN model and making the model more consistent with the high requirement of the recommended systems' timeliness(4)Optimization and application of rDNN model.By studying the effect of positive and negative samples ratio and different activation functions on the validity of the model,it confirms that the ratio to positive and negative samples is about 3,the activation function was Relu function,and when the number of the hidden layer is 3,rDNN could exert its maximum utility.In addition,the rDNN model can learn consumer behavior patterns,give more reliable prediction results,it has higher application value.