Purpose: identify potential issues that reviewers might point out, and resolve them before submission.
The method is called Adversarial writing: review your own paper yourself, consider all the questions a reviewer might ask, and resolve them one by one.
1. Three things are needed for a paper to be accepted
- Contribution is sufficient—needs to include several of the following: Novel task / Novel pipeline / Novel pipeline module / Novel design choices / New experimental findings / New insights
- Experimental results are better than previous methods.
- Ablation studies and comparison experiments are sufficient.
2. Five-aspect rejection checklist
Go through them one by one, and you will know whether this paper should be rejected.For each item, give "yes/no + evidence"; do not gloss over it.
1. Is the technical contribution sufficient (does the paper bring new knowledge to readers)
- 1.1 Are the failure cases you want to solvevery common？
- 1.2 Is the proposed techniquealready well-explored and is the performance improvement brought by this techniquepredictable / well-known？
- 1.3 Is the technique relatively straightforward?
If 1.1 or 1.2 is hit, the problem lies in topic selection rather than writing—go back to
research-ideationand check whether it falls into the first two of the "four situations." At this point, changing the writing cannot save it.
2. Is the paper writing clear
- 2.1 Does the Introductionclearly describe the paper's contributions？
- 2.2 Does the Pipeline figureclearly describe the pipeline and technical contributions？
- 2.3 Does itlack technical details and is it irreproducible？
- 2.4 Does every method module have its motivation written？
For fixes, see
paper-revision。
3. Are the experimental results good enough
- 3.1 Is it onlya little bit？
- 3.2 Although it is better than previous methods,are the results themselves still not good enough / not impressive enough？
4. Is the experimental testing sufficient
- 4.1 Is itmissing ablation studies？
- 4.2 Is itmissing important baselines? Is itmissing important evaluation metrics？
- 4.3 Is the data too simple, making it impossible to prove whether the method really works?
5. Is the method design reasonable
- 5.1 Is the experimental setting unrealistic？
- 5.2 Does the methodhave technical flaws and look unreasonable？
- 5.3 The method'srobustness: does it need hyperparameters tuned for each scenario?
- 5.4 While the new method design brings benefits, does itintroduce stronger limitations, causing the new method'snet benefit to be negative？
Criterion for 5.4:A paper should improve one metric and not obviously harm other metrics.
3. Two hard requirements that must be checked
1. All claims must hold and be supported by experiments.
All claims in the paper (especially those in the abstract and introduction) need to be error-free and supported by experiments. Otherwise, some reviewers will reject the paper outright on that basis.
Approach: list every assertion in the abstract and introduction, and annotate each one with which table / figure / section supports it. For those with no support, either add experiments or rewrite them.
2. The related work section must discuss the papers most relevant to your own method.
If you omit the most relevant work, some reviewers will reject the paper outright on that basis.
IV. Output Template
Use this format for review output.Every item must include a location and evidence—— "The introduction is not written clearly enough" is of no use to the author; "when §1, paragraph 3 introduces the technical challenge, it jumps directly to our method without explaining why recent method 2 exists" is useful.
## 自评审：<论文标题> <日期>
**总判**：<接收 / 边缘 / 会被拒> 主要风险：<一句话>
### 命中项（按严重度排序）
| # | 类别 | 位置 | 问题 | 证据 | 修复动作 | 需要多久 |
|---|---|---|---|---|---|---|
| 1 | 4.1 缺 ablation | §5.2 | core contribution B 没有单独的 ablation | 表 3 只有 A+B 和 baseline | 补 A-only 一行 | 2 天实验 |
| 2 | 2.4 缺 motivation | §3.3 第一段 | 模块 C 直接给了设计，没说为什么需要它 | | 补一句 problem-driven 开头 | 30 分钟 |
### 五方面逐条判定
1. 技术贡献是否足够：<是/否 + 依据> 1.1 ☐ 1.2 ☐ 1.3 ☐
2. 写作是否清楚：2.1 ☐ 2.2 ☐ 2.3 ☐ 2.4 ☐
3. 实验效果是否够好：3.1 ☐ 3.2 ☐
4. 实验测试是否充分：4.1 ☐ 4.2 ☐ 4.3 ☐
5. 方法设计是否合理：5.1 ☐ 5.2 ☐ 5.3 ☐ 5.4 ☐
（☐ = 未命中即通过；☒ = 命中，需在上表出现）
### Claim 审计
| Abstract/Intro 里的 claim | 支撑它的 table/figure/section | 成立？ |
|---|---|---|
| | | |
### 时间可行性
距截稿 <N> 天。上表中来得及修的：<#…>；来不及的：<#…，建议怎么处理>
The last section is very important:Reporting an issue three days before the deadline that requires two weeks of experiments is as good as not reporting it. For items that cannot be fixed in time, give a fallback plan (rewrite the claim, add it to limitations, turn strong assertions into qualified assertions), rather than leaving the author anxious.
V. When Reviewing Others' Papers
Use exactly the same five-aspect checklist and the same table, go through it item by item, and you can reach a conclusion. The only difference is that the "time feasibility" section is not needed.
VI. Make Self-Review Part of the Paper
Add a self-review question list at the end of the paper, ask questions according to the five aspects above, then revise the paper based on these questions. After revising, go through it again until there are no hits left on the list.
Pursuing perfectionism is a very important way to ensure paper quality.
VII. Beyond "Not Being Rejected"
The checklist only ensures not being rejected. To make the paperreceive good reviews, there is another layer:
Make the paper beautiful and visually appealing, so that the first impression makes people feel this paper is high-class. Three levers: a good-looking teaser figure / pipeline figure; good-looking tables and result figures; neat formatting.
To make the workhave impact(not just be accepted): push the algorithm onto harder data, build demos for downstream research communities, see research-experiment;
for impact criteria at the topic-selection level, see research-ideation。