A Usage-Centric Take on Intent Understanding in E-Commerce

Anonymous

TL;DR: We revisit intent understanding from a usage-centric perspective as a natural language reasoning task, aiming to detect superficially distinct kinds of products useful for common usage intents

Abstract: Identifying and understanding user intents is a pivotal task for E-Commerce. Despite its popularity, intent understanding has not been consistently defined or accurately benchmarked. In this paper, we focus on predicative user intents as ``how a customer uses a product'', and pose intent understanding as a natural language reasoning task, independent of product ontologies. Through topology analysis, we highlight two weaknesses in FolkScope, the SOTA E-Commerce Intent Knowledge Graph, precluding it from effectively reasoning about user intents and recommending diverse useful products. Following these observations, we propose a product recovery benchmark to isolate intent understanding abilities from confounders. We verify the identified weaknesses, and discuss future directions for intent understanding.

Paper Type: short

Research Area: NLP Applications

Contribution Types: Model analysis & interpretability, Data resources, Position papers

Languages Studied: English