讲座题目：AI-Generated Personalized Review Summaries to Reduce Product Returns: A Language Model Based Framework and Experiment（人工智能产生的个性化评论总结用于降低产品退货率：基于语言模型的框架设计和实地实验）

主讲人：谢迎 教授 University of Texas at Dallas

讲座时间：2024年10月17日10:30

讲座地点：学院319

主办单位：武汉大学经济与管理学院市场营销与旅游管理系

讲座主要内容：

The increasing rate of product returns presents a considerable challenge for online retailers. Online reviews provide valuable insights into user experiences that can potentially decrease product returns. However, the sheer abundance of reviews can overwhelm potential customers and hinder their access to useful information. This research proposes a framework that generates a personalized review summary for each customer-product pair, aiming to reduce return rates while preserving customer conversion and retention. We employ a multi-objective optimization approach to rank attributes and opinions within product reviews in a two-dimensional space. This information is then fed into a global attention-based abstractive summarizer as part of a small language model developed to produce personalized review summaries. The effectiveness of our framework was tested in a randomized, controlled digital experiment, which demonstrated a significant decrease in return rates across different product categories when customers were presented with personalized review summaries. These results offer valuable insights for online retailers seeking to reduce product return and develop optimal personalization strategies for review summaries.

产品退货率的上升给在线零售商带来了巨大的挑战。在线评论提供了有关用户体验的宝贵见解，可能会减少产品退货。然而，大量的评论可能会让潜在客户不知所措，阻碍他们获取有用的信息。这项研究提出了一个框架，为每个客户-产品对生成个性化的评论摘要，旨在降低退货率，同时保持客户转化率和保留率。我们采用多目标优化方法在二维空间中对产品评论中的属性和意见进行排名。然后将这些信息输入到基于全局注意力的抽象摘要器中，作为为生成个性化评论摘要而开发的小型语言模型的一部分。我们在一个随机控制的数字实验中测试了我们框架的有效性，结果表明，当向客户提供个性化评论摘要时，不同产品类别的退货率显著下降。这些结果为寻求减少产品退货和制定评论摘要最佳个性化策略的在线零售商提供了宝贵的见解。

主讲人简介：

谢迎教授是德克萨斯大学达拉斯分校（University of Texas at Dallas）纳维恩·金达尔管理学院（Naveen Jindal School of Management）的营销学教授。研究重点是通过量化方法探究信息在消费者决策制定中的作用，并为企业、监管机构及其他利益相关方提供启示，关注与消费者学习、社会影响、社交媒体、内容营销、创作者经济以及数字平台相关的主题。研究成果经常发表于顶尖的营销和管理类期刊上，包括 Journal of Marketing Research, Marketing Science, Management Science, and MIS Quarterly.

Ying Xie is Professor of Marketing at the Naveen Jindal School of Management, University of Texas at Dallas. Her research focuses on using quantitative methods to study the role of information in consumer decision making and derive implications for firms, regulators, and other stakeholders. She is especially interested in topics related to consumer learning, social influence, social media, content marketing, creator economy, and digital platforms. Her work has frequently appeared in top marketing and management journals including Journal of Marketing Research, Marketing Science, Management Science, and MIS Quarterly.