对于临床决策支持的大型语言模型中检索增强生成的高级分片的比较评估
Cesar Abraham Gomez-Cabello1, Srinivasagam Prabha1, Syed Ali Haider1
1Division of Plastic Surgery, Mayo Clinic, Jacksonville, FL 32224, USA.
Bioengineering (Basel, Switzerland)
|November 27, 2025
概括
通过将文本段与逻辑主题对齐,自适应块化显著提高了检索增强生成 (RAG) 的准确性和相关性,以支持临床决策,优于固定长度方法.
科学领域:
- 自然语言处理自然语言处理.
- 医疗信息学 医疗信息学
背景情况:
- 检索增强生成 (RAG) 质量对文档细分非常敏感.
- 固定长度块可以通过分割概念或引入噪音来降低RAG性能.
研究的目的:
- 评估命题,语义和自适应块化策略是否提高RAG的准确性和相关性,以支持临床决策.
- 将这些策略与固定长度基线进行比较.
主要方法:
- 通过使用临床知识库和Gemini 1.0 Pro开发了四个RAG管道,仅在块化策略 (适应性,命题,语义,固定长度) 上有所不同.
- 对每个管道提出了30个术后鼻整形问题.
- 测量结果包括医学准确性,临床相关性和检索指标 (精度,回忆,F1).
主要成果:
- 适应性分块产生了最高的准确度 (87%) 和相关性 (93%),明显超过了基线 (50%的准确性,68%的相关性;p=0.001).
- 与基线 (0.17,0.40,0.24) 相比,自适应块化也实现了优异的检索指标 (精度0.50,回忆0.88,F1 0.64).
- 比起基线,命题和语义块改进了指标,但效果不如自适应块.
结论:
- 将文本块与逻辑主题边界对齐,特别是通过自适应块,大大提高了RAG的准确性和相关性.
- 这种方法提供了一个模型不可知的方法来提高基于LLM的临床决策支持的安全性和实用性,而不会改变语言模型.
更多相关视频
05:47Evidence-based Knowledge Synthesis and Hypothesis Validation: Navigating Biomedical Knowledge Bases via Explainable AI and Agentic Systems
Published on: June 13, 2025
1.3K
07:50A Metadata Extraction Approach for Clinical Case Reports to Enable Advanced Understanding of Biomedical Concepts
Published on: September 20, 2018
16.4K
相关概念视频
Chunking
365
Chunking is a powerful cognitive technique that improves short-term memory retention by organizing information into smaller, more manageable units. The brain, limited by working memory capacity, can more easily process and store information when it is divided into "chunks" rather than presented as discrete, unrelated elements. Chunking is especially useful when dealing with large amounts of information, such as numerical sequences, words, or complex ideas.
The principle behind chunking...
The principle behind chunking...
365
Chunking and Rehearsal in Sensory Memory
546
Improving short-term memory can be achieved through techniques like chunking and rehearsal. Chunking involves organizing information into larger, more manageable units. This technique is particularly useful for information that exceeds the typical memory span of between five and nine items. For instance, logging into an online account with a password like "ta89vq0179gz" involves grouping letters and numbers into three chunks—ta89, vq01, and 79gz. It makes large amounts of...
546
