A Novel Approach for Perceptions of Physician Decision-Making and Latent Topic Refinement in Large Language...

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Medical Informatics

Background:

• Medical dialog systems (MDSs) leverage advanced medical dialog generation (MDG) for expert responses.
• Current MDSs struggle with generic responses, semantic imprecision, and inaccurate topic extraction.

Purpose of the Study:

• To design a novel model addressing challenges in medical dialog generation.
• To improve semantic precision and dialog topic extraction in MDSs.

Main Methods:

• Proposed the TRL-HMIE model, integrating GPT-3 and transformer reinforcement learning (RL).
• Developed a conversation-topic classifier and a multi-head attention mechanism for precise information extraction.
• Utilized RL with a reward fusion mechanism for optimized doctor response generation.

Main Results:

• The TRL-HMIE model demonstrated a 6.07% improvement over benchmark models.
• Achieved enhanced accuracy in heterogeneous medical information extraction.
• Generated responses with improved semantic precision and topic relevance.

Conclusions:

• The TRL-HMIE model effectively addresses key limitations in current medical dialog generation.
• The proposed enhancements lead to more accurate and precise medical dialogue systems.
• This work contributes to advancing the capabilities of AI in medical communication.