Navigating the Shortcut Maze: A Comprehensive Analysis of Shortcut Learning in Text Classification by Language Models
Yuqing Zhou, Ruixiang Tang, Ziyu Yao, Ziwei Zhu.Published in Findings of EMNLP, 2024
We propose a shortcut framework that categorizes shortcuts in text classification into three main types: occurrence, style, and concept. Then, develop a benchmark aligned with this framework to systematically evaluate model robustness against each shortcut type.