evo-citation-checker

Citation Verification Skill

Identifies fake/hallucinated citations in BibTeX files using multi-signal verification.

Verification Strategy

1. DOI-first triage: Check DOIs via CrossRef API and doi.org resolution
2. Provenance signals: DBLP biburl/bibsource and ACL Anthology URLs indicate real entries
3. DOI registrant analysis: Suspicious prefixes like 10.1234, 10.5678 indicate fabrication
4. Semantic Scholar search: For entries without DOIs, search by title
5. Red flag analysis: Generic titles, fake journal names, metadata inconsistencies

Key Fake Citation Indicators

• DOI returns 404 from CrossRef AND doi.org
• DOI registrant code is a placeholder (10.1234, 10.5678, etc.)
• Journal/venue name doesn't exist in any index
• Publisher attribution is wrong for the claimed venue
• Generic/vague title + generic author names + no provenance

Usage

import sys
sys.path.insert(0, '/app/environment/skills/evo-citation-checker/scripts')
from utils import verify_citations, validate_answer

# Run end-to-end verification
answer, details = verify_citations('/root/test.bib', '/root/answer.json')
print(answer)

# Validate output
validate_answer('/root/answer.json')

Scripts

• scripts/utils.py: All utility functions including:
  • parse_bib_file(path) - Parse BibTeX
  • clean_bibtex_title(title) - Remove BibTeX formatting
  • check_doi_crossref(doi) - Verify DOI via CrossRef
  • check_doi_resolution(doi) - Verify DOI via doi.org
  • search_semantic_scholar(title) - Search S2 by title
  • has_dblp_provenance(entry) - Check DBLP provenance
  • has_acl_anthology_url(entry) - Check ACL Anthology URL
  • is_suspicious_doi_prefix(doi) - Check for fake DOI prefixes
  • analyze_entry_suspicion(entry) - Score suspicion signals
  • verify_citations(bib_path, output_path) - End-to-end entry point
  • validate_answer(output_path) - Validate output format