Prompt Combines Paraphrase: Enhancing Biomedical “Pre-training, Prompt and Predicting” Models by Explaining Rare Biomedical Concepts

Anonymous

Abstract: Prompt-based fine-tuning for pre-trained models has been proven resultful in general domain for few-shot learning in downstream tasks. As to the biomedical domain, rare biomedical entities, which are quite ubiquitous in healthcare contexts, can affect the performance of pre-trained models, especially in low-resource scenarios. We propose a simple yet effective approach to helping models understand rare biomedical words during tuning with prompt. Experiments demonstrate that our method can achieve up to 5% improvement in biomedical tasks without any additional parameters or training steps in few-shot vanilla prompt settings.