This skill helps users start from a vague research direction and gradually complete: topic refinement → framework construction → section-by-section writing → de-AI-flavor polishing.
Please read before use：references/innovation-patterns.md(reference for innovation point types) and references/writing-templates.md(writing templates for each chapter)
Step 1: Research Direction Diagnosis and Topic Refinement
What the user provides is often a broad direction, such as "research on low-yield identification of geese." Your task is to narrow it down to an executable paper title.
1.1 Startup Questions (ask the user first, then continue)
If the user has not provided the following information, ask one by one (do not ask too many at once; prioritize the first two):
- Target scenario: Where will this research be applied? (Farm, hospital, factory, urban transportation...)
- Core technical means: What method do you plan to use? (Deep learning, sensor fusion, statistical modeling...)
- Data conditions: Is there a dataset? Is it self-built or a public dataset?
- Preferred innovation angle: Do you lean toward engineering application (system type) or academic publication (method/model type)?
- Target journal/conference: Core journal, EI, SCI, or an ordinary journal?
1.2 Topic Refinement Logic
After collecting enough information, generate 3 candidate titlesaccording to the following dimensions, and explain the applicable scenario for each:
| Title style | Template structure | Applicable situation |
|---|---|---|
| Method type | Research on [object] [task] method based on [technical means] | Leans toward algorithm innovation, submit to academic journals |
| System type | Design and implementation of [object] [task] system | Leans toward engineering, submit to engineering journals |
| Model type | Research on [object] [task] model integrating [feature A] and [feature B] | Emphasizes model construction, submit to AI/ML journals |
| Application type | Research on [object] [task] technology for [scenario] | Emphasizes application value, submit to application-oriented journals |
Note: The title must contain the three core elements of "scenario word" + "technical word" + "task word," and avoid being too broad (it cannot end with only the word "research").
Step 2: Innovation Point Identification and Research Gap Design
This is the core of whether the paper can be published. Read references/innovation-patterns.md to understand the five types of innovation point combination patterns, then for the user's direction:
2.1 Identify Usable Innovation Combinations
From the following five categories, identify the 2-3 categories that this research can cover, and provide specific explanations:
- Scenario innovation: To which new scenario has an existing technology been applied?
- Method innovation: What new model/algorithm/fusion approach is proposed?
- System innovation: Does it form a complete closed-loop system?
- Data innovation: Is a dataset for a specific scenario constructed?
- Application value innovation: Is it validated in a real environment?
A paper does not need to cover all five categories; usually a combination of 2-3 categories is enough to support a Chinese core journal paper.
2.2 Research gap sentence patterns (for writing the introduction)
Generate 3 standard statements of "existing deficiencies", with sentence templates as follows:
现有[相关技术]多侧重于[局限方向]，在[本文场景]中存在[具体问题]，缺乏[本文能提供的]。
Step 3: Overall paper framework construction
Following the six-part research route of "background phenomenon → research question → existing deficiencies → method of this paper → experimental validation → conclusion and contributions", generate a complete outline:
1. 引言
1.1 研究背景（现象驱动）
1.2 研究问题（痛点聚焦）
1.3 现有研究综述及不足
1.4 本文贡献（创新点列举）
2. 相关工作
2.1 [技术方向A]相关研究
2.2 [技术方向B]相关研究
2.3 现有方法局限性分析
3. 方法/模型设计（核心章节）
3.1 整体框架设计
3.2 [子模块1]设计
3.3 [子模块2]设计
3.4 [融合/决策]机制
4. 实验与结果分析
4.1 实验数据集与环境
4.2 评价指标
4.3 对比实验
4.4 消融实验（如有）
4.5 结果分析与讨论
5. 结论
5.1 研究总结
5.2 局限性与未来工作
Step 4: Write each chapter section by section
Read references/writing-templates.md to obtain the specific writing templates and sentence pattern library for each chapter.
Recommended writing order (do not start with the introduction!)
① 方法章节 → ② 实验章节 → ③ 结果分析 → ④ 引言与相关工作 → ⑤ 摘要与结论
The logic of this order: first write clearly what you "actually did", then work backward to frame its significance.
Step 5: De-AI-flavor polishing
This is the most easily overlooked but extremely important step. After writing the content, you must proactively counter-check the output.
5.1 Common features of AI flavor (to be proactively avoided)
| AI writing | Human writing replacement |
|---|---|
| "In summary, this paper proposes..." | State the conclusion directly |
| "It is worth noting that" | Delete it and say it directly |
| "This paper aims to..." | "Aiming at the problem of..., this paper proposes..." |
| Overly neat three-point parallelism | Replace with contrastive or progressive sentences |
| Every paragraph starts with a topic sentence | Allow entry from phenomena or data |
| Large amounts of "first... second... finally" | Switch to logical connectives: "on this basis", "furthermore" |
5.2 Specific operations for de-AI-flavor
- Add specific numbers and details: Do not write "the effect is significantly improved"; write "the accuracy increased from 82.3% to 91.7%"
- Add candid expressions of research limitations: Do not always have positive conclusions
- Use discipline-specific jargon: Each field has its own terminology habits; avoid overly generic descriptions
- Mix sentence lengths: AI tends to write sentences of equal length, while human writing has both long and short ones
- Allow vagueness and uncertainty: You can write "preliminarily validated", "indicates to a certain extent", rather than "fully proves"
5.3 Self-check list after polishing
- There are no three consecutive sentences starting with "this paper"
- There are no AI marker words such as "it is worth noting that" or "in summary"
- Numerical expressions are specific (numbers with decimal points are more natural)
- There is at least one candid expression of limitation or uncertainty
- The first sentence of the introduction is a phenomenon/data, not "with the development of..."
Appendix: Four-sentence abstract structure (for quick generation)
| Sentence position | Content | Example |
|---|---|---|
| Sentence 1 | Background + core problem | In large-scale farming, the identification of low-yield individuals relies on manual experience, which is inefficient and has a high misjudgment rate |
| Sentence 2 | The method in this paper | This paper proposes... |
| Sentence 3 | Experimental results (with numbers) | Experiments show that, compared with the baseline method, the accuracy is improved by X percentage points |
| Sentence 4 | Research value | The research results can provide technical reference for... |
The first sentence of the abstractDo notStart with "With the rapid development of...", which is the most typical feature of AI abstracts; starting directly from the problem or data is more natural.