Prediction Model Of Short-term Consumption Hotspot Based On Commodity Label

The rapid economic development has promoted the upgrading of consumption.In recent years,it has returned to people's vision that the business model of physical store represented by "new retail" in order to enhance the consumer's user experience and satisfy people's pursuit of consumer quality.The operation of physical stores faces greater pressure for storage and turnover of capital.Therefore,accurately predicting consumption hot spots is beneficial to merchants in formulating more reasonable operation strategies and reducing operating costs.Based on sales data of women's shoes in Taobao,this paper conducts research on hotspot prediction for consumption.The hotspot prediction model is designed in this paper by analyzing the characteristics of hot label and the relationship between labels.The core content of the model includes two parts.First,it is about the extraction and initialization of label since the proper label extraction is a prerequisite for prediction.In this paper,we build a custom lexicon and build a label vector to achieve label extraction,based on which the initialization of eigenvalues and the merging of synonymous labels are completed;Second,it is about the design of the combined predictor.Each label has completely different historical data at the same time so that the prediction of all labels can not done through one model.To complete the rapid modeling of many labels,in this paper,the prediction algorithm base on time series is used as a sub-algorithm of the predictor.Through the selection criteria of the unit root calibration model,the optimal parameters of the model are solved automatically.When the time series prediction algorithm changes the law of data change,the prediction error increases.In order to improve the prediction accuracy of the model,according to the periodicity of the label,the correlation analysis of label features is carried out,and the sales volume prediction algorithm based on clustering idea is designed and the algorithm can better predict the change characteristics of labels.By fitting the first derivative of the function and judging the change trend of the data,the final result of the combined predictor is output.Finally,the validity of this prediction model is verified on the real Taobao dataset.