evo-citation-scorer

Compares parsed BibTeX citation metadata against API results using fuzzy string matching and multi-factor scoring to classify citations as real or fake, then outputs the final JSON report.

SKILL.md

GitHub

evo-citation-scorer

Scores and classifies citations as real or fake.

Key Functions

  • normalize_string(text) - Normalize for comparison (lowercase, no punctuation)
  • calculate_title_similarity(title1, title2) - Token-based title similarity (0-100)
  • compute_composite_score(citation, api_result) - Multi-factor score (title 50%, author 30%, year 20%)
  • classify_citation(verification_result) - Classify as Real/Fake based on scores
  • export_fake_citations_json(fake_titles, output_file) - Write sorted JSON output

Classification Logic

  1. DOI resolves via CrossRef → Real
  2. No DOI or DOI fails → search CrossRef by title
  3. Best composite score >= 70 → Real
  4. Below 70 → Fake

Usage

import sys
sys.path.insert(0, '/app/environment/skills/evo-citation-scorer/scripts')
from utils import classify_citation, export_fake_citations_json

# After verification
status = classify_citation(verification_result)
if status == 'Fake':
    fake_titles.append(verification_result['title'])

export_fake_citations_json(fake_titles, '/root/answer.json')