20260706:新增一些文章
This commit is contained in:
41
concepts/rubric-based-evaluation.md
Normal file
41
concepts/rubric-based-evaluation.md
Normal file
@@ -0,0 +1,41 @@
|
||||
---
|
||||
title: "量规评估(Rubric-Based Evaluation)"
|
||||
created: 2026-07-02
|
||||
updated: 2026-07-02
|
||||
type: concept
|
||||
tags: [evaluation, rubric, judge, frontend]
|
||||
sources:
|
||||
- "[[verification-horizon-no-silver-bullet]]"
|
||||
---
|
||||
|
||||
# 量规评估(Rubric-Based Evaluation)
|
||||
|
||||
将整体评估分解为细粒度评分维度的结构化评估方法,减少 model-based judge 的主观偏差和不一致性。
|
||||
|
||||
## 论文中的设计
|
||||
|
||||
Qwen Team 的前端评估 rubric 将评估分解为 6 个维度,平均每任务 25.9 个 checklist 条目:
|
||||
|
||||
| 维度 | 占比 |
|
||||
|------|:---:|
|
||||
| Functional(功能正确性) | 37.7% |
|
||||
| Content(内容完整性) | 19.0% |
|
||||
| Visual(视觉质量) | 13.3% |
|
||||
| Layout(布局) | 12.9% |
|
||||
| UX(用户体验) | 9.3% |
|
||||
| Technical(技术质量) | 7.2% |
|
||||
|
||||
## 效果验证
|
||||
|
||||
- **Judge-Model 排名一致性**:跨 scorer 家族 Kendall τ ≥ 0.93
|
||||
- **Prompt 变体鲁棒**:改变 strictness、thinking level 不改变排名
|
||||
- **人机对齐**:Spearman ρ = 0.81-0.91
|
||||
|
||||
## 局限
|
||||
|
||||
静态评估无法覆盖需要运行时行为的场景(表单验证、动态路由、多页导航),这推动了 [[interactive-judge|交互式判断器]] 的发展。
|
||||
|
||||
## 参考
|
||||
|
||||
- [[verification-horizon-no-silver-bullet|论文原文]]
|
||||
- [[interactive-judge|交互式判断器]]
|
||||
Reference in New Issue
Block a user