Scholarlink

Large language models (LLMs) like GPT and PaLM have transformed natural language processing, excelling in tasks from creative writing to scientific reasoning. However, in high-stakes domains like healthcare, LLMs face limitations such as hallucinations, overconfidence, and difficulty with nuanced reasoning, especially in question answering (QA). They often struggle with datasets requiring deep reasoning or do main expertise, like in medical and legal applications, resulting in high latency and suboptimal accuracy. In this work, we propose a novel method combining iterative response refinement with relevant keyword augmentation to enhance context comprehension and output precision. Evaluations on MedQA and PubMedQA show improvements over baselines, with accuracy gains of up to 9.59%.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Question Answering Precision of Black Box LLMs Using Response Refinement and Relevant Keyword Incorporation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Parakh Chaudhary,
  • Sweta Jain,
  • Sri Khetwat Saritha

摘要

Large language models (LLMs) like GPT and PaLM have transformed natural language processing, excelling in tasks from creative writing to scientific reasoning. However, in high-stakes domains like healthcare, LLMs face limitations such as hallucinations, overconfidence, and difficulty with nuanced reasoning, especially in question answering (QA). They often struggle with datasets requiring deep reasoning or do main expertise, like in medical and legal applications, resulting in high latency and suboptimal accuracy. In this work, we propose a novel method combining iterative response refinement with relevant keyword augmentation to enhance context comprehension and output precision. Evaluations on MedQA and PubMedQA show improvements over baselines, with accuracy gains of up to 9.59%.