考虑商品组合效应的多任务需求预测模型

Multi-Task Demand Forecasting Model Considering the Commodity Combination Effect

多任务需求预测模型(MT-LR)通过商品组合效应解决短期需求预测数据不足的问题。商品组合效应指的是在订单中高频出现且具有代表性的商品组合有着相似的需求趋势。MT-LR首先通过隐狄利克雷模型(latent dirichlet allocation,LDA)学习商品的特征表达,然后通过多任务学习(multi-task learning,MTL)框架共享商品之间的销售数据。与目前主流的需求预测模型相比,MT-LR在两个真实销售数据的短期预测中有更好的表现。

Multi-Task Linear Regression(MT-LR)model solve the lack of demand data in short-term demand forecasting problem by considering the effect of commodity combination.The commodity combination effect refers to the fact that a representative combination of products that occurs frequently in orders has a similar demand trend.MT-LR first learns the feature expression of commodities through the Latent Dirichlet Allocation(LDA)model,and then shares sales data between commodities through the framework of Multi-Task Learning(MTL).Compared with the current mainstream demand forecasting model,MT-LR has better performance in two real sales data set.