| With the rapid development of the Internet industry,the e-commerce market has experienced from the rise to prosperity,but now the domestic e-commerce market is the "blue ocean" to "Red Sea",the era of large-scale standardized production has passed,focusing on refinement and intelligent production and operation to become an e-commerce Enterprises seeking to develop new growth points.The online clothing market has always occupied a large share of the e-commerce consumer market,and women’s clothing occupies an important position in the clothing category.The demand for women’s clothing is influenced by women’s social status,income level and aesthetic ability,etc.The demand for women’s clothing has shifted from covering the body to avoiding the cold to emphasizing the presentation of personality and expressing their unique style.It is particularly important for e-commerce companies to accurately analyze and predict the personalized needs of female consumers.The maturity of big data technology and data mining technology has proposed a new solution logic for personalized demand analysis and prediction in the e-commerce market.However,in practice,it remains a challenge for e-commerce companies to effectively combine business scenarios with mining technologies.In this paper,the following work is mainly done by combining YS’s operation data:First,in the analysis of personalized demand for women’s clothing in e-commerce,this paper starts from the e-commerce women’s clothing consumption chain,uses the e-commerce funnel model to quantify and analyze the important factors affecting consumer decisions in the shopping chain,and uses the dimensional disassembly method to explore the characteristics of personalized demand for women’s clothing in e-commerce from different personalized labels of women’s clothing,time and other dimensions combined with analytical tools such as offset and heat matrix.The analysis results found that each link in the e-commerce consumption chain has different degrees and directions of influence on demand generation,the funnel link conversion is complex,and the strategy applied in the link with high conversion rate has a significant effect on the improvement of demand.Ecommerce women’s personalized demand shows "small aggregation and large dispersion",fluctuating and seasonal obvious,explosive pull and other characteristics.Secondly,in terms of feature construction and feature screening,combining the results of business funnel analysis and personalized demand characteristics analysis,we summarized the feature dimensions in e-commerce demand prediction,conducted feature construction,and extracted 57 effective features.Based on RFR and XGBoost feature-weighted integrated ranking model for feature screening,25 input prediction models were screened out from 50 numerical features.The final output of the prediction models is accurate,which verifies the correctness and scientificity of the constructed features and the proposed feature screening models based on business understanding resilience.Finally,in e-commerce women’s personalized demand forecasting,a LightGBM demand forecasting model(GA-LightGBM)based on genetic algorithm optimization is proposed and comparison experiments are designed to compare the proposed model with weak learners such as multiple linear regression and decision tree,integrated learners such as RFR and XGBoost,different optimization models based on LightGBM such as RSM-LightGBM and TPE-LightGBM are compared in terms of RMSE,training time consuming,etc.It is concluded that the proposed model in this paper can have good prediction performance on a small number of feature sets,which proves the reference value of this paper in solving personalized demand analysis and prediction problems in corporate practice. |
