Machine Learning for Peace:

Digital Tools for Civic Actors

PDRI-DevLab

University of Pennsylvania

December 5, 2023

Principal Investigators: Erik Wibbels, Jeremy Springman

Data Scientists: Zung-Ru Lin, Hanling Su

Affiliates: Serkant Adiguzel, Mateo Villamizar Chaparro, Diego Romero, Rethis Togbedji Gansey, Jitender Swami

MLP: Approach


How can data help civil society?

  1. Awareness: hard data on recent events
    • Scraping reputable, local news sites to track events
    • Interactive data dashboard

  2. Planning: predictive analytics for strategic planning
    • Forecasting political events
    • 70% early warning accuracy rate

Thank You

MLP: Digital Tools

Data Production

Input: Online news

  • 300+ news sources
  • 35 languages
  • 100+ million articles

Data quality

  • Focus on reputable local sources
  • Much better coverage than extant archivers/aggregators (GDELT, Wayback, Lexis Nexis, etc.)



Output: Monthly data

  • 57+ countries
  • 2012 - last month

Successful Early Warnings

  • ~70% success rate
  • ~60 events across 25 countries

Purges:

  • Rwanda (Aug)

Legal Changes:

  • Senegal (Jul)
  • Georgia (Feb)

Arrests:

  • Kosovo (May)
  • Nicaragua (Apr)

Legal Actions:

  • Uzbekistan (Jul)
  • Kosovo (Jul)
  • Zambia (Jun)

Protests:

  • India (Mar)

Non-lethal Violence:

  • Guatemala (Jun)

Security mobilization:

  • Philippines (Apr)

Activities & Extensions

  • Distribution:
    • 2k+ unique website visitors across 102 countries (since June)
    • ~150hr/mo active app usage time
    • Mailing list of ~500
    • Data access request form (getting 1/wk)
  • Accessibility:
    • New dashboard (last week)
    • Integrating LLM-generated summaries (in development)
    • Extracting new events/info from text

mlpeace.org