Token Masking Improves Transformer-Based Text Classification

Xianglong Xu, John Bowen, Rojin Taheri

While transformer-based models achieve strong performance on text classification, we explore whether masking input tokens can further enhance their effectiveness. We propose token…