20260514:增加新内容
This commit is contained in:
36
concepts/domain-knowledge-reasoning.md
Normal file
36
concepts/domain-knowledge-reasoning.md
Normal file
@@ -0,0 +1,36 @@
|
||||
---
|
||||
title: 领域知识推理 (Domain Knowledge Reasoning)
|
||||
created: 2026-05-01
|
||||
updated: 2026-05-01
|
||||
type: concept
|
||||
tags: [benchmark, llm]
|
||||
sources: [papers/dou-cl-bench.md]
|
||||
---
|
||||
|
||||
# 领域知识推理 (Domain Knowledge Reasoning)
|
||||
|
||||
> CL-bench 的第一类上下文:模型需从上下文中**学习新领域知识并应用**来解决任务。是四种类型中最易的一类,最佳成绩 25.3%。
|
||||
|
||||
## 七个子类
|
||||
|
||||
| 子类 | 典型场景 | GPT-5.1 成绩 |
|
||||
|------|---------|-------------|
|
||||
| Finance | 新金融产品分析 | 25.2% |
|
||||
| Healthcare | 窄领域医学建议 | 21.7% |
|
||||
| Humanities | 虚构历史分析 | 23.7% |
|
||||
| Legal Advisory | 虚构法律体系断案 | 22.8% |
|
||||
| Lifestyle | 专业咨询建议 | 19.9% |
|
||||
| Management | 管理场景决策 | **34.8%** |
|
||||
| Science | 科学领域推理 | 25.8% |
|
||||
|
||||
## 核心特征
|
||||
|
||||
- 上下文提供**新的领域知识体系**(如一个虚构国家的完整法律)
|
||||
- 任务本质是**演绎推理**:将给定知识应用于具体案例
|
||||
- 相对最容易,因为知识是**显式陈述**的(不像经验发现需要归纳)
|
||||
|
||||
## 相关概念
|
||||
- [[context-learning]] — 上下文学习能力
|
||||
- [[dou-cl-bench]] — CL-bench 论文
|
||||
- [[rule-system-application]] — 规则系统应用
|
||||
- [[empirical-discovery-simulation]] — 经验发现(最难类别)
|
||||
Reference in New Issue
Block a user