摘要
研究针对在线产品销售的决策需求,结合各行业在线产品的销量影响因素及深度学习算法优势,构建了适用于在线产品的销量预测模型,并重点评估了模型在不同种类在线产品上的适应性。研究过程将全连接模型与CNN的训练结果进行了对比,证明了CNN模型的精度和泛化能力。通过选取非深度学习模型Adaboosting作为对比基线,证明CNN模型在不同类别产品下的性能优势。另外,实验得出经过无监督预训练的CNN模型在销量预测问题上更有效、适应能力更强的结论。
Targeting at decision-making requirements of online product sales, by combining the influence factors of online product sales and the advantages of deep learning, we construct a sales prediction model for all online products, and mainly evaluate the suitability of the prediction model on different kinds of online products. By comparing the training results of CNN and full-connection network, we have proved the accuracy and generalization ability of this very model. By selecting non-deep learning model Adaboosting as the comparison baseline, we have certified the performance advantage of CNN on different kinds of online products, Moreover, the results show that the CNN, through unsupervised pre-training, is more effective on predicting the sales of online products and its suitability is greater than other models.
作者
荣飞琼
郭梦飞
Rong Feiqiong;Guo Mengfei(School of Information Engineering, Lanzhou University of Finanee and Economics, Lanzhou,Gansu 730020)
出处
《西北民族大学学报（哲学社会科学版）》
CSSCI
2019年第2期15-26,共12页
Journal of Northwest Minzu University(Philosophy and Social Sciences)
基金
甘肃省软科学项目"甘肃省电子商务信用管理研究--构建基于大数据的甘肃网络供应商信用评估体系"(项目编号:17CX1ZA024)
