Literature Review

Systematic workflow for conducting comprehensive literature reviews.

When to Use

• •Starting a new research project
• •Surveying a research field
• •Writing the Related Work section of a paper
• •Identifying research gaps
• •Building a comprehensive reading list

PRISMA-Lite Workflow

This workflow adapts the PRISMA (Preferred Reporting Items for Systematic Reviews) framework for ML/AI research.

Phase 1: Define Scope

Before searching, define:

1. •Research Question: What are you trying to learn?
2. •Inclusion Criteria: What makes a paper relevant?
3. •Exclusion Criteria: What makes a paper not relevant?
4. •Time Frame: How far back to search?
5. •Search Sources: Which databases to use?

Document in literature_review.md:

## Review Scope

### Research Question
[Your research question here]

### Inclusion Criteria
- [ ] Criterion 1
- [ ] Criterion 2

### Exclusion Criteria
- [ ] Criterion 1
- [ ] Criterion 2

### Time Frame
[e.g., 2019-present]

### Sources
- [ ] Semantic Scholar
- [ ] arXiv
- [ ] Google Scholar
- [ ] ACL Anthology

Phase 2: Search

Execute systematic search using paper-finder or manual search:

1. •Primary search: Use core topic terms
2. •Secondary search: Use method/technique names
3. •Citation search: Check references of key papers

Track search queries:

## Search Log

| Date | Query | Source | Results | Notes |
|------|-------|--------|---------|-------|
| YYYY-MM-DD | "query here" | Semantic Scholar | N papers | Initial search |

Phase 3: Screening

Three-stage screening process:

Stage 1: Title Screening

• •Review all titles
• •Quick relevance judgment
• •Mark: Include / Exclude / Maybe

Stage 2: Abstract Screening

• •Read abstracts for Include/Maybe papers
• •Evaluate methodology and findings
• •Mark: Include / Exclude

Stage 3: Full-Text Screening

• •Download and read full papers
• •Verify relevance and quality
• •Extract key information

Track screening:

## Screening Results

| Paper | Title Screen | Abstract Screen | Full-Text | Notes |
|-------|-------------|-----------------|-----------|-------|
| Paper1 | Include | Include | Include | Key baseline |
| Paper2 | Maybe | Exclude | - | Different task |

Phase 4: Data Extraction

For each included paper, extract:

1. •Bibliographic: Authors, year, venue
2. •Problem: What problem is addressed?
3. •Method: What approach is used?
4. •Data: What datasets/benchmarks?
5. •Results: Key findings
6. •Limitations: Acknowledged weaknesses
7. •Relevance: How relates to our work?

Use the extraction template in assets/review_template.md.

Phase 5: Synthesis

Organize findings by theme:

1. •Identify themes: Group related papers
2. •Compare approaches: What are the differences?
3. •Find gaps: What's missing?
4. •Position work: Where does your work fit?

Output Files

literature_review.md

Main document tracking the review:

# Literature Review: [Topic]

## Review Scope
[Scope definition]

## Search Log
[Search queries and results]

## Paper Summaries
[Individual paper notes]

## Themes and Synthesis
[Grouped findings]

## Research Gaps
[Identified opportunities]

## Key Citations
[Must-cite papers for your work]

papers/ directory

Organize downloaded papers:

papers/
├── must_read/           # Relevance 3, priority reading
├── should_read/         # Relevance 2
├── reference/           # Background papers
└── README.md            # Index of all papers

Tools

Reading Large PDFs

Use the PDF chunker to split papers into smaller PDF files that can be read directly. This preserves all formatting perfectly (unlike text extraction which loses formatting).

Dependencies:

# Using uv (recommended):
uv add pypdf

# Or with pip:
pip install pypdf

How to run:

python .claude/skills/literature-review/scripts/pdf_chunker.py <pdf_path>

Options:

• •--pages-per-chunk N: Number of pages per chunk (default: 1)
• •--output-dir DIR: Output directory (default: <pdf_dir>/pages)

Output:

• •Creates PDF chunk files: <pdf_name>_chunk_001.pdf, <pdf_name>_chunk_002.pdf, etc.
• •Creates a manifest: <pdf_name>_manifest.txt listing all chunks with page ranges

Integration with screening workflow:

1. •During Phase 3 (Full-Text Screening), run the chunker on papers that need detailed review
2. •For abstract skimming: read only chunk 1 (page 1 or pages 1-3)
3. •For deep reading: read ALL chunk PDFs sequentially, writing notes after each
4. •Check the manifest to see how many chunks exist
5. •IMPORTANT: Do not skip chunks - methodology and results are in later chunks

Verify Citations

After completing the review, verify all citations are valid:

python .claude/skills/literature-review/scripts/verify_citations.py literature_review.md

Quality Checklist

• • Research question clearly defined
• • Inclusion/exclusion criteria documented
• • Multiple sources searched
• • Search queries logged
• • Screening decisions recorded
• • Key information extracted from all included papers
• • Papers organized by theme
• • Research gaps identified
• • Citations verified

References

See references/ folder for:

• •screening_guide.md: Detailed screening criteria

See assets/ folder for:

• •review_template.md: Template for literature_review.md