Preprint Open access
A Comparative Explainability Framework for DeBERTa-v3 in Zero-Shot Medical Abstract Classification
A comparative explainability framework is presented to audit DeBERTa-v3 under zero-shot classification of medical abstracts. The work addresses the disagreement problem in Explainable Artificial Intelligence, where different attribution methods produce divergent explanations for the same input and prediction. A natural …