ECOD: A Multi-modal Dataset for Intelligent Adjudication of E-Commerce Order Disputes

Liyi Chen, Shuaipeng Liu, Hailei Yan, Jie Liu, Lijie Wen, Guanglu Wan

Abstract: With the widespread popularity of e-commerce websites, online order reviews and scores help customers to choose quality products and stores. At the same time, the order dispute problem has gradually attracted attention. Specifically, bad reviews with low scores can have negative impacts on the store, where some unfair and biased reviews mislead other consumers. In order to maintain their reputation, stores usually submit responses against bad reviews. In this paper, we creatively define an intelligent adjudication task of e-commerce order disputes, which aims to judge disputes fairly based on customer reviews and store responses. Moreover, we construct a multi-modal dataset about E-Commerce Order Disputes (ECOD). It contains 6,366 pairs of multi-modal reviews and responses. And each dispute has an adjudication label annotated by business experts. We evaluate the ECOD dataset with baseline models, and analyze the difficulties and challenges in detail. We believe that the proposed dataset will not only facilitate future research in the field of dispute adjudication and multi-modal understanding, but also advance intelligent management for e-commerce websites.