literature-review

Literature Review

This skill guides literature search, organization, and review writing.

This is the most important principle and must be strictly followed:

English literature: Can be obtained through web search, but authenticity must be verified
Chinese literature: Clearly inform users to search on CNKI, and the AI provides search suggestions
All citations must be traceable and verifiable
For uncertain literature information, it is better not to include it than to fabricate it

This skill has multiple built-in literature processing scripts:

1. Literature Search (scholar_search.py)

Script location：scripts/scholar_search.py

Supported databases：PubMed, CrossRef, Semantic Scholar, arXiv

Supported output formats：

json - JSON format (default)
bibtex - BibTeX format, can be used directly for LaTeX
ris - RIS format, for EndNote/Zotero
apa - APA citation format
mla - MLA citation format
chicago - Chicago citation format
vancouver - Vancouver citation format

Usage

# 基本搜索
python scripts/scholar_search.py "deep learning transformer"

# 指定数据库
python scripts/scholar_search.py "neural network" --sources pubmed,crossref

# 年份过滤（当前是2026年，建议使用近年范围）
python scripts/scholar_search.py "machine learning" --year 2023-2026

# 输出 BibTeX 格式（用于 LaTeX 论文）
python scripts/scholar_search.py "landslide detection" --format bibtex -o refs.bib

# 输出 APA 引用格式
python scripts/scholar_search.py "attention mechanism" --format apa --limit 5

# JSON 输出（用于程序处理）
python scripts/scholar_search.py "quantum computing" --format json -o results.json

Output format examples

BibTeX format(for LaTeX):

@article{xu2024,
  title = {CAS Landslide Dataset: A Large-Scale and Multisensor Dataset},
  author = {Yulin Xu and Chaojun Ouyang and Qingsong Xu},
  journal = {Scientific Data},
  year = {2024},
  doi = {10.1038/s41597-023-02847-z},
}

APA format(for in-text citations):

Yulin Xu and Chaojun Ouyang (2024). CAS Landslide Dataset...

Characteristics of each database

Database	Rate limit	Abstract	Citation count	Applicable fields
CrossRef	High	Partial	Yes	All disciplines
PubMed	Medium	Requires additional request	No	Biomedicine
Semantic Scholar	Low*	Yes	Yes	All disciplines
arXiv	Low	Yes	No	CS/Physics/Mathematics

* Semantic Scholar recommends configuring an API Key to obtain a higher quota.

Checklist

Confirm the review topic and scope
Generate search keywords (Chinese and English)
English literature: execute search and organize results
Chinese literature: provide search strategy and wait for the user to provide it
Organize literature by topic
Generate an evidence-claim map
Annotate the citation slot for each piece of literature
Write the first draft of the review
Check the authenticity of all citations
Update plan/progress.md

0. Hard gate from literature to main text

Literature search is not the delivery endpoint. Before writing the Introduction, Related Work, or research status, the literature must be converted into an evidence-claim map:

| Source ID | Citation | Abstract-level finding | Usable fact | Supported claim | 引用位置 / citation slot | Risk |
|---|---|---|---|---|---|---|

Requirements:

Supported claim It must be a sentence that can be written into the main text, not "this literature is very relevant."
引用位置 / citation slot It must be specific to the paragraph role, such as "Introduction-P2 method lineage" or "RelatedWork-P3 FL-IDS limitations."
Each core claim must have at least 1 piece of strong supporting literature; key research gaps should be jointly supported by 2 or more pieces of literature.
Only information from titles, abstracts, DOI metadata, excerpts provided by the user, or full texts that have been read may be used.

If there is no evidence-claim map, it must not be claimed that the literature review has been completed.

I. Literature Search Guide

1.0 Script search (recommended)

Use scripts/scholar_search.py for parallel multi-database search.

Basic Usage

# 搜索所有数据库，输出 JSON
python scripts/scholar_search.py "your query" --format json -o results.json

# 指定数据库和年份
python scripts/scholar_search.py "deep learning" --sources crossref,semanticscholar --year 2023-2026

# 仅搜索 PubMed（生物医学）
python scripts/scholar_search.py "hippocampus memory" --sources pubmed --limit 20

BibTeX Output (for LaTeX papers)

# 输出 BibTeX 格式
python scripts/scholar_search.py "landslide detection" --format bibtex -o refs.bib

# 直接输出到控制台
python scripts/scholar_search.py "transformer attention" --format bibtex --limit 5

BibTeX Output Example：

@article{xu2024,
  title = {CAS Landslide Dataset: A Large-Scale and Multisensor Dataset},
  author = {Yulin Xu and Chaojun Ouyang and Qingsong Xu},
  journal = {Scientific Data},
  year = {2024},
  doi = {10.1038/s41597-023-02847-z},
  url = {https://doi.org/10.1038/s41597-023-02847-z}
}

Text Citation Format

# APA 格式
python scripts/scholar_search.py "neural network" --format apa --limit 3

# MLA 格式
python scripts/scholar_search.py "machine learning" --format mla --limit 3

# Chicago 格式
python scripts/scholar_search.py "attention mechanism" --format chicago --limit 3

APA Output Example：

Yulin Xu and Chaojun Ouyang (2024). CAS Landslide Dataset: A Large-Scale
and Multisensor Dataset for Deep Learning-Based Landslide Detection.
Scientific Data. 10.1038/s41597-023-02847-z

Search Strategy Suggestions

Start searching: use CrossRef (fast, many results, has citation counts)
Precise search: add a year range filter
Domain search：
- Biomedicine → PubMed
- Computer Science/Physics/Mathematics → arXiv
- Need AI to recommend related literature → Semantic Scholar

JSON Output Example

[
  {
    "title": "Attention Is All You Need",
    "authors": ["Ashish Vaswani", "Noam Shazeer", "..."],
    "year": 2017,
    "journal": "Advances in neural information processing systems",
    "doi": "10.48550/arXiv.1706.03762",
    "citations": 100000,
    "url": "https://doi.org/10.48550/arXiv.1706.03762",
    "_source": "crossref"
  }
]

1.1 WebSearch Search (Alternative)

When the script is unavailable, use WebSearch to search.

Available databases:

Database	Features	Applicable Fields
Google Scholar	Most comprehensive	All disciplines
PubMed	Biomedical authority	Medicine, biology
IEEE Xplore	Engineering technology	Computer science, electronics
arXiv	Preprints	Physics, mathematics, CS
Semantic Scholar	AI-enhanced search	All disciplines

Search strategy:

Determine core keywords (in English)
Use Boolean operators: AND, OR, NOT
Use quotation marks for exact matching: "deep learning"
Limit the time range: prioritize the last 5 years
Sort by citation count to find high-impact literature

1.2 Chinese Literature Search

Recommended databases:

CNKI: the most comprehensive Chinese academic database
Wanfang Data: rich in dissertations
VIP: journal articles
Baidu Scholar: comprehensive search

How AI can help:

Generate search keywords
Provide search strategy suggestions
After the user provides an abstract, AI helps organize it

II. Literature Organization

2.1 Recording Literature Information

Record the following information for each piece of literature:

## 文献1
- **标题**：
- **作者**：
- **年份**：
- **期刊/会议**：
- **DOI/链接**：
- **核心观点**：
- **研究方法**：
- **主要结论**：
- **与本研究关系**：

2.2 Literature Classification

Organize by topic rather than by literature:

文献综述/
├── 理论基础/
├── 方法技术/
├── 应用研究/
└── 综述文章/

III. Review Writing

3.1 Review Structure

一、引言
   - 研究背景
   - 综述目的和范围

二、研究现状
   2.1 主题一
   2.2 主题二
   2.3 主题三

三、研究评述
   - 已有研究的贡献
   - 存在的不足
   - 研究趋势

四、研究空白与本研究定位

3.2 Writing Points

Synthesize rather than list

❌ Zhang San (2020) studied A. Li Si (2021) studied B.
✅ Regarding issue X, the academic community has mainly conducted research from three angles: Zhang San (2020) started from angle A...; while Li Si (2021) focused on aspect B...

Critical analysis

Not only introduce the research content, but also evaluate its contributions and limitations
Point out the relationships between different studies (supporting, supplementing, contradicting)

3.3 Review Writing Template

Introduce a certain research field:

[领域名称]是近年来[学科]研究的热点之一。自[开创性工作]以来，
该领域经历了快速发展，主要研究集中在[方向1]、[方向2]和[方向3]等方面。

Introduce representative research:

[作者]（[年份]）提出了[方法/理论]，该研究[主要贡献]。
实验结果表明，[主要发现]。然而，该方法存在[局限性]。

Point out the research gap:

尽管已有研究在[方面]取得了显著进展，但在[具体问题]方面仍存在不足。
具体而言，[问题1]尚未得到充分探讨，[问题2]缺乏系统性研究。

IV. Citation Formats

Chinese (GB/T 7714-2015)

[1] 作者. 文章题目[J]. 期刊名, 年份, 卷(期): 页码.
[2] 作者. 书名[M]. 出版地: 出版社, 年份: 页码.

English (APA 7th)

Author, A. A., & Author, B. B. (Year). Title. Journal, Volume(Issue), pages.

V. Dialogue Templates

Scenario 0: Use a script to search literature (recommended)

Execute search command：

python scripts/scholar_search.py "搜索关键词" --sources crossref,semanticscholar --year 2020-2024 --limit 20 --format json

Output example：

"Let me use the academic search script to help you find literature.

Search query：[关键词] Data source：CrossRef, Semantic Scholar Year range：2020-2024

Search results(sorted by citation count):

# Title Author Year Journal Citation count DOI

1 [...] [...] 2023 [...] 156 [DOI]

2 [...] [...] 2022 [...] 89 [DOI]

Do you need me to further retrieve detailed information for a particular piece of literature?"

#	Title	Author	Year	Journal	Citation count	DOI
1	[...]	[...]	2023	[...]	156	[DOI]
2	[...]	[...]	2022	[...]	89	[DOI]

Scenario 1: The user needs English literature (WebSearch method)

"Let me help you search for relevant English literature.

Search keywords: [generated keywords]

Search results：

[Literature title] - [Author] ([Year])

Journal: [Journal name]

Core content: [brief description]

DOI: [link]

Suggestion: Among these works, [Literature X] is most relevant to your research."

Scenario 2: The user needs Chinese literature

"Chinese academic literature needs to be searched in databases such as CNKI. Let me help you prepare a search strategy.

Suggested search keywords：

Subject terms: [Keyword 1], [Keyword 2]

Combined search: [Keyword 1] AND [Keyword 2]

Search steps：

Visit https://www.cnki.net/

Select "Advanced Search"

Enter the above keywords

Filter conditions: core journals/CSSCI, past 5 years

Please paste the abstracts of the literature you find for me, and I will help you organize and analyze them."

Scenario 3: The user provided literature abstracts

"Received. Let me help you organize this literature.

Literature organization：

No. Author Year Core viewpoint Relationship to this study

1 [Author] [Year] [Viewpoint] [Relationship]

Review suggestion: This literature can be organized from the following perspectives...

Do you need me to help you draft a review paragraph?"

No.	Author	Year	Core viewpoint	Relationship to this study
1	[Author]	[Year]	[Viewpoint]	[Relationship]

VI. Citation checking and verification

6.1 Citation verification (required)

Use verification skill for strict verification:

Claim	Verification method	Insufficient
Citation exists	CrossRef API confirms DOI	"Looks correct"
Citation format is correct	Run format check script	Visual inspection
Author information is accurate	Search the original source	"Should be correct"

Verification script：

# 验证 DOI 是否存在
curl -s "https://api.crossref.org/works/10.1000/doi123"

# 验证 BibTeX 文件
python scripts/scholar_search.py "your query" --format bibtex --output refs.bib

6.2 PDF literature parsing

When the user provides a PDF file, use scripts/pdf_parser.py to extract content:

# 提取 PDF 文本
python scripts/pdf_parser.py paper.pdf --output paper_text.txt

# 提取结构和摘要
python scripts/pdf_parser.py paper.pdf --sections --abstract --json paper_info.json

# 总结 PDF 内容
python scripts/pdf_parser.py paper.pdf --summarize

Output content：

Metadata (title, author, page count)
Abstract
IMRaD sections (Introduction, Methods, Results, Discussion)
Automatic summary

6.3 Citation checklist

Must check:

All cited references actually exist (verified via DOI)
Author, year, and title information is accurate
Citation format is consistent
In-text citations correspond one-to-one with the reference list
Each citation has context explaining its relationship to this study