🔎 Citation Verifier

Verify the validity of citations in scientific and academic papers.
Upload a PDF/DOCX or paste a URL or DOI, and the app:

• extracts the paper’s References/Bibliography (or uses public APIs for DOI-only),
• searches for each citation (preferring DOI, else exact title),
• buckets results into ✅ Exact, ⚠️ Close, and ❌ Not found,
• shows clickable links for Exact/Close matches,
• displays a live progress bar (extraction count + per-citation status),
• and lets you download a Markdown report (paper metadata + abstract + grouped citations).

---

1) What it does

• Input sources: file upload (PDF/DOCX), URL (publisher/arXiv/DOI page), or DOI only (e.g., 10.1038/...).
• Extraction: finds the References/Bibliography section and structures citations (authors, title, year, DOI…).
• Verification:
  • uses the DOI when present; otherwise searches exact titles;
  • classifies results as Exact, Close, or Not found;
  • includes links for Exact/Close.
• Reporting: one-click Markdown report with paper metadata (title, authors, DOI, abstract) and all citations grouped by outcome.
• Resilience: DOI fetch falls back to Crossref/OpenAlex/Semantic Scholar metadata/references when publishers block bots (403).

---

2) How to use (with example)

1. Open the app (see the Hugging Face link below).
2. In the “About” box, review the quick steps.
3. Provide one of the following:
   • Upload: drag-and-drop a .pdf or .docx.
   • URL: paste a paper page (e.g., https://www.pnas.org/doi/10.1073/pnas.2507345122).
   • DOI: paste just the DOI (e.g., 10.1101/515643).
4. Click Verify Citations.
5. Watch the progress row:
   • “Extracting citations…”, then “Found N citations”,
   • “Searching citations… (k/N)” as each verification completes.
6. Browse the three tabs:
   • Exact matches (N) – verified citations with a link.
   • Close matches (M) – likely matches; check manually.
   • Not found (K) – possible errors or missing items.
7. Click Download report (Markdown) to save a full summary.

Example
Paste the DOI 10.1101/515643 (bioRxiv). Even if the publisher blocks direct PDF download, the app will pull references via Crossref/OpenAlex and proceed with verification.

---

3) Live app (Hugging Face)

👉 Open the Citation Verifier on Hugging Face
(Replace with your actual Space URL if different.)

---

4) Install & run locally

Minimum requirements

• Python ≥ 3.12
• macOS / Linux / Windows
• Internet connection (for verification + DOI/metadata lookups)

Quick start

# 1) (optional) create a fresh virtual environment
python -m venv .venv
# Windows: .\.venv\Scripts\activate
# macOS/Linux: source .venv/bin/activate

# 2) Upgrade pip
python -m pip install -U pip

# 3) Install dependencies
pip install -U gradio PyPDF2 python-docx pandas requests beautifulsoup4

# Optional (improves success rate against 403 walls):
pip install cloudscraper

# If you’re using the OpenAI Agents SDK for the search agent:
# (Make sure you’re using the correct 'agents' package)
pip install openai-agents

# 4) (optional) set environment variables
# For better open-access PDF discovery when publishers block bots:
# PowerShell (Windows):
setx UNPAYWALL_EMAIL "[email protected]"
# macOS/Linux (bash/zsh):
export UNPAYWALL_EMAIL="[email protected]"

# You will also need a .env file containing you OpenAI API key:
OPENAI_API_KEY=sk-your-key
[email protected]

# 5) Run the app
python app.py

Notes

• If you see TensorFlow/Gym errors, you likely installed the wrong agents package (an RL library). Uninstall agents and install openai-agents instead.
• For .doc (legacy Word) or scanned PDFs, convert to .docx / text-based PDF or run OCR first for best results.

---

Update history

v0.6 — 2025-09-09

• Added DOI-only input with Crossref/OpenAlex/Semantic Scholar fallback to avoid 403s.
• Strengthened URL fetching and added publisher heuristics (e.g., PNAS, bioRxiv).

v0.5

• New Clear Input Fields button; resets inputs and status/progress.
• Startup reset clears all tabs and counts before new runs.

v0.4

• Compact custom progress bar with text next to it; shows “Found N citations” and per-citation “(k/N)”.
• Tabbed interface with dynamic counts and clickable links for Exact/Close.

v0.3

• Downloadable Markdown report (paper metadata + abstract; grouped citations with links).
• Dark/light theme hero header box.

v0.2

• URL ingestion with PDF preference; HTML fallback if no PDF found.

v0.1

• Initial release: upload PDF/DOCX, extract references, verify citations, bucket results.

---

About

Developed by Cornell Genomics Innovation Hub (2025).
If you use this tool in your lab/classroom, a shout-out is appreciated!