利用大语言模型模拟人类认知过程实现专家级医学问答
计算与语言
2023-10-18 v1 人工智能
神经与进化计算
摘要
为应对医疗领域对先进临床问题解决工具的迫切需求，我们提出BooksMed，一种基于大语言模型(LLM)的新颖框架。BooksMed独特地模拟人类认知过程以提供基于证据且可靠的回应，利用GRADE（推荐分级、评估、制定与评价）框架有效量化证据强度。为恰当评估临床决策，需要一种临床对齐且经过验证的评估指标。作为解决方案，我们提出ExpertMedQA，一个由开放式、专家级临床问题构成的多专科临床基准，并由多元化医学专业人员验证。通过要求对最新临床文献的深入理解和批判性评估，ExpertMedQA严格评估LLM性能。BooksMed在多种医疗场景中优于现有最先进模型Med-PaLM 2、Almanac和ChatGPT。因此，模拟人类认知阶段的框架可成为为临床询问提供可靠且基于证据回应的有用工具。
引用
@article{arxiv.2310.11266,
title = {Emulating Human Cognitive Processes for Expert-Level Medical Question-Answering with Large Language Models},
author = {Khushboo Verma and Marina Moore and Stephanie Wottrich and Karla Robles López and Nishant Aggarwal and Zeel Bhatt and Aagamjit Singh and Bradford Unroe and Salah Basheer and Nitish Sachdeva and Prinka Arora and Harmanjeet Kaur and Tanupreet Kaur and Tevon Hood and Anahi Marquez and Tushar Varshney and Nanfu Deng and Azaan Ramani and Pawanraj Ishwara and Maimoona Saeed and Tatiana López Velarde Peña and Bryan Barksdale and Sushovan Guha and Satwant Kumar},
journal= {arXiv preprint arXiv:2310.11266},
year = {2023}
}