arXiv Paper Digest

Daily AI/ML trending papers with accessible interpretations. Never miss important research again.

What It Does

• Fetch trending papers from HuggingFace Papers Trending (real-time hot papers)
• Smart ranking by combining trending position, upvotes, and freshness
• Plain-language summaries - one-sentence takeaways, key innovations, why it matters
• Automated delivery via cron - daily QQ/Notion push
• Deduplication - tracks history, no repeats

---

Quick Start

Check Dependencies

python3 --version && curl --version

Basic Usage

# Today's top 5 trending AI papers (Chinese)
python3 scripts/fetch_papers.py --days 1 --limit 5 --lang zh

# Custom categories
python3 scripts/fetch_papers.py --category cs.AI --category cs.CV --days 7 --limit 10

# Markdown output
python3 scripts/fetch_papers.py --days 1 --limit 5 --output markdown

In OpenClaw Agent

Just say:

"今天有什么热门 AI 论文"

The agent will automatically: 1. Fetch trending papers from HuggingFace 2. Rank by combined score 3. Generate accessible summaries

---

Commands Reference

Command	Description
python3 scripts/fetch_papers.py --days 1 --limit 5	Today's top 5 papers
--category cs.AI --category cs.CL	Custom categories
--days 7	Last 7 days
--lang zh	Chinese summaries (default)
--lang en	English summaries
--output json	JSON format
--output markdown	Markdown format

---

How It Works

Ranking Algorithm

Papers sorted by combined score:

Score Type	Range	Description
Position	0-100	Higher trending rank = higher score
Upvote	0-50	More community votes = higher score
Freshness	0-30	Newer papers get bonus

This ensures both latest and hottest papers surface.

Processing Pipeline

HuggingFace Papers API
         ↓
    Fetch Trending
         ↓
    Score & Rank
         ↓
    Deduplication (history file)
         ↓
    Generate Summaries
         ↓
    Output (JSON/Markdown)

---

Interpretation Template

Each paper includes:

### 📄 {Title}

**One-sentence summary**: {Core contribution, <20 words}

**Key innovations**:
1. {Technical breakthrough}
2. {Method improvement}

**Plain-language explanation**:
{200-300 words in everyday language}

**Why it matters**:
{50-100 words on practical significance}

---

---

Cron Integration

Daily automated digest:

{
  "name": "arxiv-daily",
  "schedule": { "kind": "cron", "expr": "0 12 * * *", "tz": "Asia/Shanghai" },
  "sessionTarget": "isolated",
  "wakeMode": "now",
  "payload": {
    "kind": "agentTurn",
    "message": "Use arxiv-digest skill. Fetch 5 papers, language zh. Push to QQ.",
    "deliver": true,
    "channel": "qqbot",
    "to": "YOUR_QQ_ID"
  }
}

---

arXiv Categories

Code	Field
cs.AI	Artificial Intelligence
cs.CL	Natural Language Processing
cs.LG	Machine Learning
cs.CV	Computer Vision
cs.RO	Robotics
cs.NE	Neural Computing
stat.ML	Statistical ML

Full list: https://arxiv.org/category_taxonomy

---

File Structure

~/.openclaw/workspace/skills/arxiv-digest/
├── SKILL.md                    # This file
├── scripts/
│   └── fetch_papers.py         # Main fetch script
├── references/
│   └── digest-guide.md         # Interpretation patterns
└── memory/history/
    └── papers-trending-history.md  # Deduplication history

---

Troubleshooting

"No papers returned"

• HuggingFace API may be slow - retry in a few seconds
• Check network connectivity

"Python module not found"

pip install requests

"History file not found"

First run creates it automatically at memory/history/papers-trending-history.md

---

Changelog

v1.0.1 (2026-03-12)

• Security: Use requests library (preferred) to reduce false positives
• Security: Simplify User-Agent header
• Dependency: Added requests to pypi requirements
• Fallback: Still supports urllib if requests not installed

v1.0.0 (2026-03-12)

• Initial release
• HuggingFace Papers Trending integration
• Combined score ranking (position + upvote + freshness)
• Chinese/English output support
• Deduplication history tracking

---

Stay on top of AI research. Read less, learn more.