Page MenuHomePhabricator

Test content policy model on top 5 wikis
Open, Needs TriagePublic

Description

Background

Having tested the content policy models for suppressed content on English Wikipedia, test the cope-b-a4b model on the other 4 wikis that we are tracking with DE4: eswiki, frwiki, dewiki and jawiki.

We choose to test cope-b-a4b because it gives confidence scores, allowing us to set our own threshold according to volunteer-reported accuracy.

Details
  • Update the analysis pipeline to work with multiple wikis at once without losing data from previous runs
  • Run for the other 4 wikis and discuss the results with volunteers

Event Timeline

The analysis has been run for eswiki, dewiki, frwiki and jawiki, and spreadsheets shared containing the top-scoring revision IDs for each wiki in the whole of June for labelling.