Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Satisfactory Medical Consultation based on Terminology-Enhanced Information Retrieval and Emotional In-Context Learning

Kaiwen Zuo, Jing Tang, Hanbing Qin, Binli Luo, Ligang He, Shiyan Tang

TL;DR

The paper addresses the gap between current LLM-based medical consultations and professional standards by introducing a retrieval-augmented framework that fuses Terminology-Enhanced Information Retrieval (TEIR) with Emotional In-Context Learning (EICL). TEIR components—Terminology Detector, Terminology Memory, and Enhanced Sentence Generation—coupled with EICL's Emotionally Attuned Generation and Gradient-Guided Refinement (optimized via suffix-tuning) extend context handling and embed emotional awareness into responses, formalized with objectives such as $\max_{\phi} P(y | x; \theta; \phi)$ and the softmax-based terminology scoring $\hat{p}_n = map(p_n) = \frac{\exp(p_n)}{\sum_{i=1}^N \exp(p_i)}$. The approach is validated on a large Chinese dataset of 803,564 online medical consultations spanning 12 departments, showing improvements in BLEU and ROUGE metrics over five baselines and higher simulated patient satisfaction (e.g., 36.83% in AMT studies), with ablations confirming the significance of TEIR and EICL. The findings suggest that TEIR+EICL can extend model context, improve diagnostic relevance, and provide emotionally nuanced guidance, potentially enhancing patient satisfaction and informing medical education and digital health services. Formally, the training objective and retrieval scoring mechanisms underpin the improved coherence, relevance, and emotional alignment of generated consultations, signaling a practical pathway toward more empathetic, proactive online medical advisory systems.

Abstract

Recent advancements in Large Language Models (LLMs) have marked significant progress in understanding and responding to medical inquiries. However, their performance still falls short of the standards set by professional consultations. This paper introduces a novel framework for medical consultation, comprising two main modules: Terminology-Enhanced Information Retrieval (TEIR) and Emotional In-Context Learning (EICL). TEIR ensures implicit reasoning through the utilization of inductive knowledge and key terminology retrieval, overcoming the limitations of restricted domain knowledge in public databases. Additionally, this module features capabilities for processing long context. The EICL module aids in generating sentences with high attribute relevance by memorizing semantic and attribute information from unlabelled corpora and applying controlled retrieval for the required information. Furthermore, a dataset comprising 803,564 consultation records was compiled in China, significantly enhancing the model's capability for complex dialogues and proactive inquiry initiation. Comprehensive experiments demonstrate the proposed method's effectiveness in extending the context window length of existing LLMs. The experimental outcomes and extensive data validate the framework's superiority over five baseline models in terms of BLEU and ROUGE performance metrics, with substantial leads in certain capabilities. Notably, ablation studies confirm the significance of the TEIR and EICL components. In addition, our new framework has the potential to significantly improve patient satisfaction in real clinical consulting situations.

Satisfactory Medical Consultation based on Terminology-Enhanced Information Retrieval and Emotional In-Context Learning

TL;DR

The paper addresses the gap between current LLM-based medical consultations and professional standards by introducing a retrieval-augmented framework that fuses Terminology-Enhanced Information Retrieval (TEIR) with Emotional In-Context Learning (EICL). TEIR components—Terminology Detector, Terminology Memory, and Enhanced Sentence Generation—coupled with EICL's Emotionally Attuned Generation and Gradient-Guided Refinement (optimized via suffix-tuning) extend context handling and embed emotional awareness into responses, formalized with objectives such as and the softmax-based terminology scoring . The approach is validated on a large Chinese dataset of 803,564 online medical consultations spanning 12 departments, showing improvements in BLEU and ROUGE metrics over five baselines and higher simulated patient satisfaction (e.g., 36.83% in AMT studies), with ablations confirming the significance of TEIR and EICL. The findings suggest that TEIR+EICL can extend model context, improve diagnostic relevance, and provide emotionally nuanced guidance, potentially enhancing patient satisfaction and informing medical education and digital health services. Formally, the training objective and retrieval scoring mechanisms underpin the improved coherence, relevance, and emotional alignment of generated consultations, signaling a practical pathway toward more empathetic, proactive online medical advisory systems.

Abstract

Recent advancements in Large Language Models (LLMs) have marked significant progress in understanding and responding to medical inquiries. However, their performance still falls short of the standards set by professional consultations. This paper introduces a novel framework for medical consultation, comprising two main modules: Terminology-Enhanced Information Retrieval (TEIR) and Emotional In-Context Learning (EICL). TEIR ensures implicit reasoning through the utilization of inductive knowledge and key terminology retrieval, overcoming the limitations of restricted domain knowledge in public databases. Additionally, this module features capabilities for processing long context. The EICL module aids in generating sentences with high attribute relevance by memorizing semantic and attribute information from unlabelled corpora and applying controlled retrieval for the required information. Furthermore, a dataset comprising 803,564 consultation records was compiled in China, significantly enhancing the model's capability for complex dialogues and proactive inquiry initiation. Comprehensive experiments demonstrate the proposed method's effectiveness in extending the context window length of existing LLMs. The experimental outcomes and extensive data validate the framework's superiority over five baseline models in terms of BLEU and ROUGE performance metrics, with substantial leads in certain capabilities. Notably, ablation studies confirm the significance of the TEIR and EICL components. In addition, our new framework has the potential to significantly improve patient satisfaction in real clinical consulting situations.

Paper Structure

This paper contains 18 sections, 2 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Keyword Retrieval Generation
  4. Template Engineering and Sentiment Annotation Generation
  5. Methodology
  6. Terminology-Enhanced Information Retrieval (TEIR)
  7. Emotional In-Context Learning
  8. Experiment
  9. Experiment setup
  10. Datasets
  11. Baselines.
  12. Experimental settings.
  13. Experiment result
  14. Ablation Study
  15. Ablation of TEIR
  16. ...and 3 more sections

Figures (4)

  • Figure 1: An example of model responses, which involves retrieving information about dental treatments, processing inquiries about these treatments, and identifying the emotional tone of the information presented.
  • Figure 2: The overall pipeline of Terminology-Enhanced Information Retrieval (TEIR).
  • Figure 3: The geometric diagram illustrates the emotion-influenced iterative generation process. Representations for the event schema, the document, and the $I$-th discrete demonstration are symbolized by $\mathbf{I}_e$, $\mathbf{I}_x$, and $\mathbf{I}_d^{(i)}$ correspondingly.
  • Figure 4: 1v1 battle between different models and ours