paper-review

学术论文评审技能

概述

此技能为顶级计算机科学会议和期刊的学术论文提供全面的同行评审。评审会依据原创性、技术质量、实验验证、清晰度和重要性等标准评估论文。

支持的输入格式

1. PDF 文件：直接分析编译后的论文
2. LaTeX 文件：分析源 .tex 文件

评审流程

步骤 1：收集评审背景信息

开始评审前，使用 AskUserQuestion 收集以下信息：

1. 输出语言：询问评审应使用中文还是英文
2. 目标会议或期刊：询问会议或期刊名称（例如，“NeurIPS 2026”“CVPR 2026”“ICML 2026”“TPAMI”）

交互示例：

AskUserQuestion with two questions:
1. "What language should the review be written in?"
   - Options: "English", "Chinese (中文)"
2. "What is the target conference or journal for this paper?"
   - Options: "NeurIPS", "CVPR", "ICML", "ICLR", "Other (please specify)"

步骤 2：提取论文内容

对于 PDF 文件：

import pdfplumber

with pdfplumber.open(pdf_path) as pdf:
    full_text = ""
    for page in pdf.pages:
        full_text += page.extract_text() + "\n\n"

对于 LaTeX 文件：

# Read main .tex file and any \input or \include files
import re

def read_latex_file(filepath):
    with open(filepath, 'r', encoding='utf-8') as f:
        content = f.read()

    # Find \input{} and \include{} commands
    includes = re.findall(r'\\(?:input|include)\{([^}]+)\}', content)

    # Read included files recursively
    for inc in includes:
        if not inc.endswith('.tex'):
            inc += '.tex'
        inc_path = os.path.join(os.path.dirname(filepath), inc)
        if os.path.exists(inc_path):
            inc_content = read_latex_file(inc_path)
            content = content.replace(f'\\input{{{inc}}}', inc_content)
            content = content.replace(f'\\include{{{inc}}}', inc_content)

    return content

步骤 3：分析论文结构

提取并识别以下关键章节：

• 摘要
• 引言
• 相关工作
• 方法/方案
• 实验/结果
• 讨论
• 结论
• 参考文献

步骤 4：开展全面评审

从以下维度评估论文：

1. 原创性与新颖性（评分：X/10）

• 问题是否新颖，或方法是否具有创新性？
• 与现有工作有何不同？
• 贡献是否清晰陈述？
• 信心等级：1-5（1=不熟悉，5=专家）

2. 技术质量与可靠性（评分：X/10）

• 方法在技术上是否可靠？
• 主张是否有理论或证据支持？
• 是否存在技术错误或漏洞？
• 信心等级：1-5

3. 实验验证（评分：X/10）

• 实验是否全面且设计合理？
• 基线是否恰当，比较是否公平？
• 结果是否具有统计显著性？
• 是否包含消融研究？
• 信心等级：1-5

4. 清晰度与呈现（评分：X/10）

• 论文是否行文流畅、结构清晰？
• 概念是否解释清楚？
• 符号使用是否一致？
• 是否存在语法或语言问题？
• 信心等级：1-5

5. 重要性与影响（评分：X/10）

• 这项工作对该领域有多重要？
• 有何潜在影响？
• 是否开辟了新的研究方向？
• 信心等级：1-5

6. 可复现性（评分：X/10）

检查以下方面：

• 代码是否可用（文中提及或已提供）
• 数据集说明及其可用性
• 超参数详情
• 实验设置是否完整
• 计算资源需求
• 信心等级：1-5

7. 视觉元素质量

评估：

• 图：清晰度、相关性、质量
• 表：完整性、格式、可读性
• 示意图：准确性、实用性
• 图注：信息是否充分

8. 伦理与局限性

评估：

• 是否讨论局限性
• 潜在的负面社会影响
• 伦理考量
• 更广泛影响声明（若会议或期刊有要求）

步骤 5：生成总体评估

按加权平均计算总分：

• 原创性：20%
• 技术质量：25%
• 实验验证：20%
• 清晰度：15%
• 重要性：15%
• 可复现性：5%

给出建议：

• 强烈接收（9-10）：论文出色，明确建议接收
• 接收（8-8.9）：论文良好，应予接收
• 弱接收（7-7.9）：达到门槛，倾向接收
• 临界（6-6.9）：难以定夺，接收或拒绝皆有可能
• 弱拒绝（5-5.9）：未达门槛，倾向拒绝
• 拒绝（4-4.9）：明确建议拒绝，存在重大问题
• 强烈拒绝（1-3.9）：存在根本性缺陷，强烈建议拒绝

步骤 6：撰写评审文档

创建 review-{paper-name}-{timestamp}.md，并采用以下结构：

# Paper Review: [Paper Title]

**Venue:** [Conference/Journal Name]
**Review Date:** [YYYY-MM-DD HH:MM]
**Review Language:** [Chinese/English]
**Reviewer:** Claude AI Assistant

---

## Overall Recommendation

- **Decision:** [Strong Accept / Accept / Weak Accept / Borderline / Weak Reject / Reject / Strong Reject]
- **Overall Score:** [X.X/10]
- **Confidence:** [5-Expert in this area / 4-Knowledgeable / 3-Familiar / 2-Somewhat familiar / 1-Not familiar]

---

## Summary

[2-3 paragraph summary covering:
- What the paper is about
- Main contributions
- Key strengths and weaknesses
- Overall assessment]

---

## Strengths

- [Strength 1]
- [Strength 2]
- [Strength 3]
- ...

---

## Weaknesses

- [Weakness 1]
- [Weakness 2]
- [Weakness 3]
- ...

---

## Detailed Evaluation

### 1. Originality & Novelty
**Score:** X/10 | **Confidence:** X/5

[Detailed assessment of novelty, comparison with related work, clarity of contributions]

### 2. Technical Quality & Soundness
**Score:** X/10 | **Confidence:** X/5

[Assessment of technical correctness, theoretical foundations, methodology soundness]

### 3. Experimental Validation
**Score:** X/10 | **Confidence:** X/5

[Evaluation of experimental design, baselines, results, ablation studies, statistical significance]

### 4. Clarity & Presentation
**Score:** X/10 | **Confidence:** X/5

[Assessment of writing quality, organization, notation, figures, readability]

### 5. Significance & Impact
**Score:** X/10 | **Confidence:** X/5

[Evaluation of importance, potential impact, relevance to community]

### 6. Reproducibility
**Score:** X/10 | **Confidence:** X/5

[Assessment of code availability, dataset details, hyperparameters, experimental setup]

### 7. Visual Elements Quality

[Analysis of figures, tables, and diagrams - clarity, completeness, appropriateness]

### 8. Ethics & Limitations

[Discussion of ethical considerations, limitations, broader impact, potential negative consequences]

---

## Questions for Authors

1. [Question 1]
2. [Question 2]
3. [Question 3]
...

---

## Suggestions for Rebuttal

[Constructive guidance on how authors might address the main concerns raised in this review. Focus on the most critical issues that could change the recommendation if addressed.]

---

## Minor Issues

- [Typo/formatting issue 1]
- [Reference issue 2]
- [Minor suggestion 3]
...

---

## Detailed Comments by Section

### Abstract
[Comments on abstract]

### Introduction
[Comments on introduction]

### Related Work
[Comments on related work section]

### Methodology
[Comments on methodology]

### Experiments
[Comments on experiments]

### Conclusion
[Comments on conclusion]

---

**End of Review**

步骤 7：生成改进计划

创建 improvements-{paper-name}-{timestamp}.md，其中包含分类整理的待办事项：

# Paper Improvement Plan

**Paper:** [Paper Title]
**Generated:** [YYYY-MM-DD HH:MM]

This document provides a structured improvement plan based on the review. Items are categorized by priority and type.

---

## 🔴 Critical Issues (Must Address)

These issues significantly impact the paper's acceptance chances and must be addressed.

- [ ] **[Issue 1 Title]**
  - **Location:** [Section/Page]
  - **Problem:** [Description of the issue]
  - **Suggested Fix:** [How to address it]
  - **Estimated Effort:** [High/Medium/Low]

- [ ] **[Issue 2 Title]**
  - **Location:** [Section/Page]
  - **Problem:** [Description]
  - **Suggested Fix:** [Solution]
  - **Estimated Effort:** [High/Medium/Low]

---

## 🟡 Major Improvements (Strongly Recommended)

These improvements would significantly strengthen the paper.

- [ ] **[Improvement 1 Title]**
  - **Location:** [Section/Page]
  - **Current State:** [What's there now]
  - **Suggested Enhancement:** [What to add/change]
  - **Expected Impact:** [How this helps]
  - **Estimated Effort:** [High/Medium/Low]

- [ ] **[Improvement 2 Title]**
  - **Location:** [Section/Page]
  - **Current State:** [Description]
  - **Suggested Enhancement:** [Enhancement]
  - **Expected Impact:** [Impact]
  - **Estimated Effort:** [High/Medium/Low]

---

## 🟢 Minor Suggestions (Nice to Have)

These are polish items that would improve the paper but aren't critical.

- [ ] **[Suggestion 1]** - [Brief description]
- [ ] **[Suggestion 2]** - [Brief description]
- [ ] **[Suggestion 3]** - [Brief description]

---

## 📝 Writing & Presentation

- [ ] **Grammar and Language**
  - [List specific issues with line numbers if possible]

- [ ] **Notation and Terminology**
  - [Inconsistencies or unclear notation]

- [ ] **Figure and Table Improvements**
  - [Specific suggestions for visual elements]

---

## 🔬 Experimental Enhancements

- [ ] **Additional Experiments**
  - [What experiments to add and why]

- [ ] **Baseline Comparisons**
  - [Missing baselines to include]

- [ ] **Ablation Studies**
  - [What ablations would be valuable]

---

## 📚 Related Work & Citations

- [ ] **Missing References**
  - [Important papers to cite]

- [ ] **Related Work Discussion**
  - [Gaps in related work coverage]

---

## 🎯 Priority Roadmap

### Phase 1: Critical Fixes (Before Resubmission)
1. [Critical issue 1]
2. [Critical issue 2]
3. [Critical issue 3]

### Phase 2: Major Improvements (Strengthen Paper)
1. [Major improvement 1]
2. [Major improvement 2]
3. [Major improvement 3]

### Phase 3: Polish (If Time Permits)
1. [Minor suggestion 1]
2. [Minor suggestion 2]
3. [Minor suggestion 3]

---

## 📊 Estimated Impact on Review Score

If all critical issues and major improvements are addressed:
- **Current Overall Score:** X.X/10
- **Potential Score After Improvements:** Y.Y/10
- **Expected Recommendation Change:** [Current] → [Potential]

---

**End of Improvement Plan**

评审语气指南

始终保持建设性且平衡的语气：

1. 具体明确：尽可能提供具体示例和行号
2. 公平公正：既肯定优点，也指出不足
3. 着眼建设：提出可执行的建议，而不只是批评
4. 专业严谨：保持尊重、客观的措辞
5. 鼓励支持：肯定出色的工作及其潜力

示例用法

# User invokes the skill
/paper-review path/to/paper.pdf

# Or with LaTeX
/paper-review path/to/paper.tex

输出文件

该技能会在当前目录下生成两个文件：

1. review-{paper-name}-{timestamp}.md - 完整评审
2. improvements-{paper-name}-{timestamp}.md - 改进计划

注意事项

• 开始评审前，务必通读整篇论文
• 对于 LaTeX 文件，通过遵循 \input 和 \include 命令来处理多文件项目
• 根据论文篇幅调整评审深度（会议论文与期刊论文）
• 考虑特定会议的要求（例如，NeurIPS 重视可复现性，CVPR 重视实验结果）
• 留意计算机科学论文中的常见问题：过度宣称、基线不足、符号不清晰、缺少消融实验

常见计算机科学会议的评审标准

• NeurIPS：高度重视可复现性、代码提交和理论严谨性
• CVPR/ICCV/ECCV：关注实验结果、视觉质量和基准测试表现
• ICML：兼顾理论与实验，并关注统计显著性
• ICLR：开放评审流程、可复现性和代码可用性
• ACL/EMNLP：语言质量、语言学分析和数据集详情
• SIGIR/WWW：实际适用性、可扩展性和用户研究

高效评审的技巧

1. 先读摘要和结论：掌握整体内容
2. 浏览图表：了解实验设置
3. 阅读引言：了解研究动机和贡献
4. 深入研究方法：检查技术严谨性
5. 分析实验：确认论点是否有充分证据支持
6. 检查相关工作：确保论文定位恰当
7. 审视写作质量：指出清晰度方面的问题
8. 考虑可复现性：检查是否缺少必要细节