Communication unBreakdown

Enhancing targeting for community organizing using supervised and reinforcement learning

Description

Communication unBreakdown is focused on using technology and data science to help community organizers reach specific populations for an empathy-based citizen engagement program known as “Deep Canvassing.”

Leveraging a close relationship with People’s Action, voter demographic data, and responses from former and active Deep Canvass campaigns, we apply supervised machine learning and reinforcement learning techniques for our MVP, “Dynamic and Responsive Targeting System,” or DARTS, to find undecided voters for the 2020 U.S. Presidential Election, and subsequently, the Georgia Senate runoff.

The MVP was rolled out on October 20 for People's Action.

Techniques

  • supervised learning
  • reinforcement learning (multi-arm bandits)
  • feature engineering
  • Python package development

Tools

  • scikit-learn
  • Spark
  • AWS
  • PyPI
  • Flask
  • D3

Outcome

  • Developed and released a Python package, DARTS, for using multi-arm bandits under delayed feedback scenarios
  • Increased targeting effectiveness for People's Action by over 33%.

More Information

More information can be found at the following links:

Project Page: Berkeley iSchool

Project Website: https://unbreak.info/

Python Package: Python Package Index

Python Package GitHub Repository: github/darts

Project Repository: github/unbreak.info

Presentation Slides: Google Docs