Abstract: We introduce a new regression model for time series forecasting in cross-border e-commerce domain. We focus on industry sale prediction and hot product prediction. E-commerce products contain many attributes, and to fully utilize these features for better results, our model employs a novel dual layer regression architecture to improve the generalization. The first layer is a time series regression model which aims at capturing correlation between the history data and future data. The second layer is a features regression model which can enhance the relationship of features and target value in each time points. To verify the effectiveness of our proposed model, we establish two cross-border e-commerce datasets about imported lipsticks and shoes. Then we conduct extensive experiments for industry sale prediction and hot product prediction. The experimental performance demonstrates that our proposed model achieves impressive results compared to other strong baselines and the precision of hot product prediction reached 90%.
Authors: Wangda Luo (Harbin Institute of Technology (Shenzhen), China); Hang Su (Harbin Institute of Technology, Shenzhen, China); Yuhan Liu (Harbin Institute of Technology, Shenzhen & Chinese Nankai University, China); Ruifeng Xu (Harbin Institute of Technology, Shenzhen, China)
Email: firstname.lastname@example.org, email@example.com, firstname.lastname@example.org, email@example.com