Back to Details

research-paper-writing

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

research-paper-writing

研究论文写作

Use this skill when the problem is not just "write better prose", but "turn research work into a defensible paper." The focus is contribution framing, evidence ordering, experiment coverage, and reviewer-facing clarity.
当你面临的问题不只是“写出更好的文字”，而是“把研究工作转化为逻辑严谨可支撑的论文”时使用本技能。核心聚焦于贡献梳理、论据排布、实验覆盖度以及面向审稿人的内容清晰度。

When to use this skill

适用场景

  • Drafting or rewriting an abstract around concrete claims and measurable evidence
  • Structuring an introduction so motivation, gap, method, and contributions land quickly
  • Turning notes or code results into a method section with reproducible detail
  • Designing experiment, ablation, and error-analysis sections
  • Writing related work that positions the paper instead of listing references
  • Tightening a rebuttal or reviewer response under strict word limits
  • 围绕具体论点和可量化论据撰写或重写摘要
  • 搭建引言结构，让研究动机、现有研究空白、方法、贡献能快速被读者接收
  • 将笔记或代码结果转化为包含可复现细节的方法章节
  • 设计实验、消融实验和误差分析章节
  • 撰写能突出论文定位而非单纯罗列参考文献的相关工作章节
  • 在严格的字数限制下打磨审稿回复或回应内容

Instructions

使用指南

Step 1: Lock the paper contract

第一步：明确论文核心约定

Before writing, define:
  • the core problem in one sentence
  • the single strongest contribution
  • the minimum evidence required to defend that contribution
  • the target venue and its format constraints
If any of those are fuzzy, fix them first. Weak framing leaks into every section.
动笔前先定义清楚：
  • 一句话概括核心问题
  • 唯一的最强核心贡献
  • 支撑该贡献所需的最低限度论据
  • 目标投稿会议/期刊及其格式约束
如果以上任何一项模糊不清，先解决这些问题。模糊的定位会渗透到论文的每一个章节。

Step 2: Write the abstract from claims, not chronology

第二步：从论点出发撰写摘要，而非按时间顺序

Use this order:
  1. problem and stakes
  2. gap in existing work
  3. proposed method
  4. strongest quantitative results
  5. scope or implication
Keep numbers concrete. Replace vague phrases like "significant improvement" with metric + benchmark + margin.
使用以下结构：
  1. 问题及其重要性
  2. 现有研究的空白
  3. 提出的方法
  4. 最亮眼的量化结果
  5. 适用范围或影响
保留具体数字。将“显著提升”这类模糊表述替换为「指标+基准测试+提升幅度」的具体形式。

Step 3: Build the introduction as a reviewer funnel

第三步：把引言打造成审稿人漏斗

Structure the introduction in five moves:
  1. why the problem matters
  2. why existing approaches fall short
  3. what your method changes
  4. what the evidence shows
  5. bullet contributions
Contribution bullets should be specific and testable, not marketing copy.
按照五个步骤搭建引言结构：
  1. 阐述问题的重要性
  2. 说明现有方法的不足
  3. 你的方法带来了哪些改变
  4. 论据证明了什么
  5. 分点列出贡献
贡献要点应当具体且可验证，而非营销话术。

Step 4: Make the method reproducible

第四步：确保方法可复现

The method section should answer:
  • what inputs and outputs exist
  • what modules or stages the system contains
  • what training or optimization objective is used
  • what implementation choices materially affect results
Use equations only where they clarify behavior. If a paragraph can be replaced by a precise algorithm box or table, do that.
方法章节应当回答以下问题：
  • 有哪些输入和输出
  • 系统包含哪些模块或阶段
  • 使用了什么训练或优化目标
  • 哪些实现选型会对结果产生实质性影响
仅在能清晰说明行为的地方使用公式。如果一段文字可以用精准的算法框图或表格替换，就替换。

Step 5: Treat experiments as the proof section

第五步：将实验视作举证章节

Cover at least:
  • main benchmark results
  • ablations for the claimed mechanism
  • comparison to strong baselines
  • qualitative or failure analysis when helpful
  • efficiency or cost if the method claims practicality
Each subsection should map back to one contribution claim.
至少覆盖以下内容：
  • 核心基准测试结果
  • 针对所提出机制的消融实验
  • 与强基线方法的对比
  • 必要时的定性分析或失败案例分析
  • 如果方法宣称具备实用性，需补充效率或成本相关内容
每个子章节都应当对应回一个贡献论点。

Step 6: Write rebuttals with evidence first

第六步：撰写审稿回复时优先放论据

For each reviewer concern:
  1. restate the concern precisely
  2. answer directly in one sentence
  3. add concrete evidence
  4. say what will change in the camera-ready version, if applicable
Do not become defensive. Remove throat-clearing and persuasion language that is not backed by evidence.
针对每一条审稿人意见：
  1. 精准重述意见内容
  2. 用一句话直接给出答复
  3. 补充具体论据
  4. 说明如果适用，终稿中会做出哪些修改
不要带有防御性态度。删除没有论据支撑的铺垫和说服性表述。

Examples

示例

Example 1: Abstract rewrite

示例1：摘要重写

Input:
  • notes on a diffusion model paper
  • benchmark table
  • target venue: CVPR
Output:
  • a 150-200 word abstract with problem, gap, method, results, and impact
输入：
  • 一篇diffusion model论文的相关笔记
  • 基准测试表格
  • 目标投稿会议：CVPR
输出：
  • 150-200词的摘要，包含问题、研究空白、方法、结果和影响

Example 2: Experiment plan

示例2：实验规划

Input:
  • draft method section
  • three claimed contributions
Output:
  • experiment matrix listing datasets, baselines, ablations, metrics, and figure/table owners
输入：
  • 方法章节草稿
  • 三项宣称的贡献
输出：
  • 实验矩阵，列出数据集、基线方法、消融实验、指标以及图表负责人

Best practices

最佳实践

  1. Every major claim should have a matching figure, table, or ablation.
  2. Do not bury the best result in the middle of a paragraph.
  3. Use consistent terminology for modules, datasets, and metrics throughout the paper.
  4. Prefer short, information-dense sentences over long narrative transitions.
  5. If a result is mixed, state the boundary clearly instead of overselling.
  1. 每一个核心论点都应当有对应的图表或消融实验支撑
  2. 不要把最好的结果藏在段落中间
  3. 全文对模块、数据集、指标使用统一的术语表述
  4. 优先使用简短、信息密度高的句子，而非冗长的叙事过渡
  5. 如果结果好坏参半，清晰说明适用边界，不要过度夸大

References

参考资料

  • Peng Sida, research paper writing notes
  • NeurIPS, ICLR, CVPR author guidelines
  • Peng Sida 研究论文写作笔记
  • NeurIPS、ICLR、CVPR作者指南