Paper Compare

Compare academic research papers side-by-side to identify similarities, differences, and research gaps. Use when user wants to compare 1-5 papers via DOIs, URLs, search queries, or PDF files. Supports mixed input types. Outputs both comparison table and detailed narrative summary.

KeXu9@kexu9

Install

openclaw skills install @kexu9/paper-compare

Paper Compare

Compare academic papers side-by-side with structured tables and detailed narrative analysis.

The Paper Comparison Reasoning Framework

text

┌─────────────────────────────────────────────────────────────┐
│  PAPER COMPARISON THINKING                                   │
├─────────────────────────────────────────────────────────────┤
│  1. INTERPRET  → What papers? What comparison goal?        │
│  2. RETRIEVE   → Fetch metadata, abstracts, full text     │
│  3. ANALYZE    → Extract across 10 dimensions              │
│  4. SYNTHESIZE → Build narrative, find gaps, score quality │
│  5. VALIDATE   → Check completeness, deliver              │
└─────────────────────────────────────────────────────────────┘

Decision Tree: Input Processing

text

USER INPUT
    │
    ├── 1 paper ──→ Single Paper Summary
    │       └── Skip comparison, show full summary
    │
    ├── 2-5 papers ──→ Full Comparison
    │       └── Proceed with 10 dimensions
    │
    ├── >5 papers ──→ Ask to Narrow
    │       └── "Please narrow to 2-5 for meaningful comparison"
    │
    ├── DOI ──→ Fetch via crossref/semantic scholar
    │       └── https://api.crossref.org/works/{doi}
    │
    ├── URL ──→ Fetch via web_fetch
    │       └── Extract title, authors, abstract
    │
    ├── Search query ──→ Search first
    │       └── Use web_search, present top 3, CONFIRM before proceeding
    │
    └── PDF file ──→ Extract text first
            └── Use pdf skill, then extract metadata

Decision Tree: Comparison Angle

text

WHAT IS THE COMPARISON ABOUT?
    │
    ├── Same topic, different methods ──→ 
    │       └── Focus: methodology differences, results comparison
    │
    ├── Same method, different domains ──→
    │       └── Focus: adaptation, performance across domains
    │
    ├── Evolution over time ──→
    │       └── Focus: improvements, what changed, SOTA progression
    │
    ├── Competing approaches ──→
    │       └── Focus: trade-offs, when to choose which
    │
    └── Complementary papers ──→
            └── Focus: how they combine, gaps each fills

Self-Check: After Identifying Angle

Does my analysis focus on the right aspects?
Will this help the user make a decision?

Step 1: Interpret the Request

What to Clarify

Question	Why It Matters
Which papers?	Need exact references
What goal?	Learning? Research? Writing?
What comparison angle?	Focus analysis appropriately

Self-Check: Before Starting

Do I have all paper references?
Do I understand what user wants to learn?
Is the number of papers appropriate (1-5)?
What's the comparison angle?

Step 2: Retrieve Papers

Retrieval Strategy

Input Type	Method	Source
DOI	API	crossref, semantic scholar
URL	web_fetch	arXiv, IEEE, PubMed
Search	web_search → web_fetch	Find, then confirm
PDF	pdf skill	Extract text
History	memory_search	Prior comparisons

Quality Priority

text

Must have:
├── Title
├── Authors
├── Year
├── Venue
├── Abstract (for methodology + results)

Nice to have:
├── Full text (for limitations)
├── Code/data links
├── Citation count (see below)

Citation Count

Use Semantic Scholar API:

text

https://api.semanticscholar.org/graph/v1/paper/{doi}?fields=citationCount

Self-Check: After Retrieval

Did I get the abstract?
Can I determine the methodology?
Are there any papers with missing critical info?
Did I get citation counts?

Step 3: Analyze (10 Dimensions)

Core Dimensions (Always Include)

#	Dimension	What to Extract
1	Title	Full title
2	Authors	All authors, first author highlighted
3	Year	Publication year
4	Venue	Journal/Conference
5	Research Question	What problem do they solve?
6	Methodology	Approach, techniques used
7	Dataset	What data did they use?
8	Results	Key findings with numbers
9	Limitations	What do they acknowledge?
10	Code & Data	Links to artifacts?

Decision: What If Missing?

text

Missing dimension:
    │
    ├── Abstract missing ──→ Note "Unable to analyze methodology"
    │
    ├── Results missing ──→ Note "Results not available in metadata"
    │
    ├── Limitations missing ──→ Note "Not specified" (don't infer)
    │
    └── Dataset unclear ──→ Note "Not clearly specified"

Step 4: Synthesize

Quality Scoring

Evaluate each paper:

Factor	Score	Notes
Venue Quality
- Top-tier (NeurIPS, ICML, ICLR, Nature, Science)	⭐⭐⭐
- Good (AAAI, IJCAI, CVPR, EMNLP, IEEE)	⭐⭐
- Other	⭐
Citations
- 100+	⭐⭐⭐	Highly cited
- 10-100	⭐⭐	Well-known
- <10	⭐	Recent or niche
Code Available
- Yes, official	⭐⭐⭐
- Yes, community	⭐⭐
- No	⭐
Data Available
- Yes	⭐⭐⭐
- No	⭐

Overall Quality: Sum stars (higher = more established)

Comparison Table Structure

text

| Dimension | Paper A | Paper B | ... |
|-----------|---------|---------|-----|
| Title | ... | ... | ... |
| Authors | ... | ... | ... |
| Year | ... | ... | ... |
| Venue | ... | ... | ... |
| Research Question | ... | ... | ... |
| Methodology | ... | ... | ... |
| Dataset | ... | ... | ... |
| Results | ... | ... | ... |
| Limitations | ... | ... | ... |
| Code & Data | ... | ... | ... |
| Quality Score | [⭐⭐⭐] | [⭐⭐] | ... |

Narrative Synthesis Template

Structure:

text

## Overview
[What problem each paper addresses - high-level]
[Comparison angle: what are we comparing?]

## Methodology Comparison
[Compare techniques - are they compression-based? architecture-based?
 What's the key algorithmic difference?
 How does the comparison angle affect this?]

## Results Analysis
[Quantitative results - specific numbers, metrics
 Performance comparison - trade-offs mentioned
 Which paper wins on what?]

## Limitations
[What each paper acknowledges - be honest about gaps]
[What's NOT covered that might matter]

## Research Gaps
[What's MISSING across ALL papers]
[What's not yet explored]
[Potential future directions]

## Quality Assessment
[Paper A: ⭐⭐⭐ - Why]
[Paper B: ⭐⭐ - Why]
[Note any concerns]

Step 5: Structured Verdict

Decision Matrix

text

| If You Need... | Choose | Why |
|----------------|--------|-----|
| [Best performance] | Paper [X] | [Reason] |
| [Easiest to implement] | Paper [X] | [Reason] |
| [Latest method] | Paper [X] | [Reason] |
| [Most cited/reliable] | Paper [X] | [Reason] |
| [Code available] | Paper [X] | [Reason] |

Final Recommendation

markdown

## Verdict

**For [user's goal]:**

- **Best overall:** [Paper X] — [key reason]
- **Best for implementation:** [Paper Y] — [key reason]  
- **Best for research depth:** [Paper Z] — [key reason]

**My recommendation:** [Paper X] because [specific reason matching user's goal]

**If you're unsure:** Start with [Paper X] for [reason], then explore [Paper Y] if you need [different aspect].

Self-Check: Before Delivering

Did I answer the user's original question?
Did I identify the comparison angle?
Are all 10 dimensions covered?
Is quality scored?
Is verdict actionable?

Step 6: Validate & Deliver

For Single Paper (1 only)

Output:

text

## Paper Summary

**Title:** [title]
**Authors:** [authors]
**Year:** [year]
**Venue:** [venue]

### Research Question
[What problem they address]

### Methodology
[Brief description]

### Key Results
[With numbers]

### Limitations
[What they acknowledge]

### Code & Data
[Links or "Not specified"]

### Citation Count
[If available]

### Quality Score
[⭐⭐⭐]

For Comparison (2-5 papers)

Deliver:

Comparison Angle — What we're comparing and why
Comparison Table — All 10 dimensions + quality
Narrative Summary — 6-section synthesis
Quality Assessment — Scored factors
Structured Verdict — Decision matrix + recommendation

Edge Cases to Note

Situation	How to Handle
Different fields	Warn: "Comparing CS vs Biology papers"
Very different years	Note: "2010 vs 2024 — comparison may be unfair"
Preprint	Note: "Preprint — not peer-reviewed"
Conflicting results	Note: "Paper A claims X, Paper B claims Y"

Error Handling

If Retrieval Fails

text

FETCH FAILS
    │
    ├── DOI not found ──→ Check DOI format, try search
    │       └── "DOI not found. Did you mean...?"
    │
    ├── URL inaccessible ──→ Try alternative source
    │       └── e.g., arXiv → semantic scholar
    │
    ├── Search returns nothing ──→ Try different keywords
    │       └── "No papers found for [query]. Try...?"
    │
    └── PDF extraction fails ──→ Note "Unable to extract"
            └── Can still use metadata if available

History (Persistence)

Save After Comparison

json

{
  "last_comparison": {
    "date": "2026-03-04",
    "user_goal": "[what user wanted to achieve - learning/research/writing/decision]",
    "papers": [
      {"title": "...", "doi": "10.xxxx/xxx"},
      {"title": "...", "url": "..."}
    ],
    "topic": "[what was compared]",
    "comparison_angle": "[same topic different methods / etc]",
    "verdict": "[which paper recommended]",
    "dimensions": {
      "methodology": "...",
      "key_difference": "..."
    }
  }
}

Load History

Read memory/paper-compare-history.json if exists
Use memory_search to find prior comparisons

Dependencies

Skill	Use For
pdf	Extract text from uploaded PDFs
web_search	Find papers by query
web_fetch	Get paper content from URLs

Quick Reference

Input	Action
1 DOI	Single summary
2 DOIs	Full comparison
arXiv URL	Fetch abstract
"search for X"	Search → confirm → proceed
Upload PDF	Extract → analyze

Summary Checklist

Notes

Always confirm before proceeding with search results
Keep comparisons focused: 2-5 papers max
Don't infer missing information — state "Not specified"
Save to history for future reference
Quality scoring helps users make informed decisions

Paper Compare

Install

Paper Compare

The Paper Comparison Reasoning Framework

Decision Tree: Input Processing

Decision Tree: Comparison Angle

Step 1: Interpret the Request

What to Clarify

Self-Check: Before Starting

Step 2: Retrieve Papers

Retrieval Strategy

Quality Priority

Citation Count

Self-Check: After Retrieval

Step 3: Analyze (10 Dimensions)

Core Dimensions (Always Include)

Decision: What If Missing?

Step 4: Synthesize

Quality Scoring

Comparison Table Structure

Narrative Synthesis Template

Step 5: Structured Verdict

Decision Matrix

Decision Matrix

Final Recommendation

Self-Check: Before Delivering

Step 6: Validate & Deliver

For Single Paper (1 only)

For Comparison (2-5 papers)

Edge Cases to Note

Error Handling

If Retrieval Fails

History (Persistence)

Save After Comparison

Load History

Dependencies

Quick Reference

Summary Checklist

Notes

Related skills