Research On User Profile Generation And E-commerce Demand Prediction Based On Deep Learning

With the rapid development of E-commerce platform and logistics industry in China,regional freight volume is growing rapidly.At the same time,the forecast of user demand and inventory control of various categories of goods have become the bottleneck factors for the expansion and development of e-commerce enterprises.With the widespread application of artificial intelligence and big data,e-commerce platforms can better statistics and analyze the data related to users and commodities to establish user portraits,determine target users and predict demand.However,the current research on the combination of e-commerce user portrait and demand prediction is not perfect,there is a lack of relevant research on the transformation of user comment data into fine-grained demand prediction factors,and there is no e-commerce user portrait label system and matching combination prediction model for demand prediction.Therefore,this paper proposes a combined prediction model based on user portrait labels as influencing factors of demand prediction.Firstly,e-commerce user portraits are constructed to subdivide users,analyze the behavioral characteristics of user groups,and construct user labels.Then,based on the multi-dimensional interaction data between e-commerce platform users and commodities.The labels of e-commerce user portraits are used to quantify and normalize different types of e-commerce data,so that it can be used as various covariables in e-commerce demand prediction to assist in predicting future e-commerce demand.This can help merchants on the e-commerce platform predict the demand for goods in the next few weeks based on the e-commerce data,reduce their inventory costs and improve their competitiveness in the e-commerce industry.The main contents of this paper are as follows:In view of the problems that it is difficult to establish feasible influencing factors of demand prediction and analyze and process user comment data in demand prediction of small-batch and low-level e-commerce data of individual stores,this paper proposes to mine the potential information of existing e-commerce data based on user portraits.By building a funnel model and RFM model such as electricity users operating model and based on the analysis of the training model comments emotion to establish ecommerce users tag and extract the statistical characteristics of the e-commerce data into user label corresponding weights,thus the user tags as demand forecasting model of variable weight sequence to the input,auxiliary predict E-commerce users demand.After the experiment with real e-commerce data,the feasibility and rationality of transforming e-commerce data into influencing factors of demand prediction based on user portraits were verified by correlation detection of user label weight sequence and e-commerce demand.In view of the combination forecast model building,this paper puts forward the adaptive evaluation index weights allocation strategy and combination forecast model to assist in the channeling of the combination forecast model to build,and this paper used the multiple evaluation index TPE super parameter optimization algorithm optimize the combination forecast model,from the demand forecast results corresponding to the MAE,RMSE and R squared three evaluation metrics,Compared with other traditional hyperparameter optimization algorithms,TPE optimization algorithm improves the accuracy and stability of e-commerce demand prediction,and has certain practical application value.