科苑经管国际学术论坛（第66期）

报告题目：Enhancing recommendations for household customers using product-customer lifecycle data

报告人：彭勃、瞿树晖  斯坦福大学工程管理系

报告时间：2018年12月20日（周四）16:00—18:00

报告地点：中国科学院大学中关村校区青年公寓7号楼401

报告摘要：With the growth of the intensely competitive e-commerce ecosystem worldwide, households have become an increasingly important customer group for the online shopping market. In general, they usually have more stable purchasing behaviors, and are more willing to recycle than individual customers. Even though household customers have different purchase and recycling behaviors, most state-of-art online shopping recommendation systems (RSs) usually don’t distinguish household and individual consumer types as different context-rich features deliberately. This neglect might result in lower prediction accuracy for the RS. To address this limitation, and improve the accuracy of the RS, we will introduce a product-customer interaction lifecycle (PCILC) model, which formed the basis of household RS. The model consists of a representation of detailed customer behaviors for their interaction with products: the intention, purchases, uses, and recycling. Our model thus enables a detailed analysis and more accurate prediction of household customer behavior. Based on the information gain analysis of this model, we show that recycle and purchase behaviors have a strong correlation. We then developed a three-step procedure RS for the household customer and implemented the algorithm. Finally, we test the performance of the three-step RS on a real-world PCILC dataset. Strikingly, our RS obtains 30% more F1 value compared with a baseline model. To our knowledge, the PCILC model is the first attempt at analyzing more comprehensive product-customer lifecycle data sets, which includes purchase, uses, and recycle information, to realistically enhance RS performance for household customers.

报告人简介：彭勃、瞿树晖，现为斯坦福大学工程管理系博士研究生。研究兴趣主要为人工智能在制造业和工程管理中的应用，相关研究成果发表在建筑工程信息领域的国际顶级期刊Automation in Construction、人工智能领域知名国际大会SAI Intelligent Systems Conference上。