Fairsearch Plugin for Elasticsearch

The Fairsearch plug-in for Elasticsearch is an implementation of the FA*IR algorithm. It enables you to apply a positive action policy in which elements are re-ranked to ensure a fair representation of minorities or disadvantaged people.

This plugin has been developed by Pere Urbón in collaboration with researchers at TU Berlin and Pompeu Fabra University, with support from a grant by the Data Transparency Lab.

Get started

• Check whether this plug-in is for you: The Theory, The Plugin
• Understand the fairness criterion applied: The Theory
• Use the plug-in to perform a re-ranking: How to use the fair-search plugin
• Advance usage, like build an M table: Advance usage

Installing

Pre-built versions can be found here. Want a build for an ES version? Follow the instructions in the README for building or create an issue. Once you’ve found a version compatible with your Elasticsearch, you’d run a command such as:

./bin/elasticsearch-plugin install https://fair-search.github.io/fair-reranker/fairsearch-1.0-es6.1.2-SNAPSHOT.zip

(It’s expected you’ll confirm some security exceptions, you can pass -b to elasticsearch-plugin to automatically install)

Heeelp!

• If you have questions or feedback, see Contact

Contents

• The Theory
  • Ranking can be biased
    • Core concepts: ranking bias
    • Real-world example: Job Search
  • The FA*IR algorithm
    • What is the fairness criterion applied by FA*IR?
    • Example
    • Corrections for multiple hypotheses testing
  • References
• The Plugin
  • What the Fairsearch plugin does
  • What the plugin does NOT
• How to use the fair-search plugin
  • Assumptions and preconditions for this example
  • How does a search looks like
• Advance usage
  • Building M tables
    • Create a new M table
    • List all stored M tables
    • Delete stored M tables
• Contact

Indices and tables

• Index
• Search Page