AAAI 2022 paper - Unifying Model Explainability and Robustness for Joint Text Classification and Rationale Extraction