Medical short text classification via Soft Prompt-tuning.

Area of Science:

• Natural Language Processing
• Medical Informatics
• Computational Linguistics

Background:

• Medical short texts (e.g., conversations, inquiries) are crucial but challenging to classify.
• Existing methods struggle with short text length, medical jargon, and feature sparsity.

Purpose of the Study:

• To develop a novel soft prompt-tuning method for enhanced medical short text classification.
• To address limitations of existing approaches in handling specialized medical data.

Main Methods:

• Introduced a soft prompt-tuning method tailored for medical short texts.
• Incorporated automatic template generation to combat short length and feature sparsity.
• Proposed two label word space expansion strategies for medical vocabulary and measures.

Main Results:

• The proposed method demonstrates effectiveness in medical short text classification.
• The approach successfully addresses key challenges including text length, sparsity, and specialized terminology.
• Experimental results indicate significant advancements in accuracy and interpretability.

Conclusions:

• The novel soft prompt-tuning method offers a promising solution for medical short text classification.
• This advancement has significant implications for understanding and utilizing medical text data.
• The method paves the way for more accurate and interpretable medical text analysis.