A Study on Question and Answer Intent Recognition in Medical Domain Based on Prompt Learning
Chiyu Shi, Chiawei Chu, Junyu Su · 2024
Intention recognition constitutes a pivotal aspect of medical Q&A systems, with its accuracy intimately tied to the precision of medical Q&A generation tasks. Addressing the challenge of data scarcity in the medical Q&A domain, our team has devised a cue learning-based model, P-BERT-IR, specifically tailored for medical Q&A intent recognition. This model adeptly identifies intent labels in scenarios with limited samples by crafting optimal cue templates. Comparative experiments against the benchmark model within the IMCS dataset have demonstrated a notable enhancement in the F1 value, ranging from 2.24% to 5.79%. Furthermore, in experiments involving small sample sizes, the model's performance exhibits minimal degradation as the number of training samples decreases.