Page MenuHomePhabricator

WE 1.3.4 Roll out Revert Risk Filters to Wikis that don't have damaging/goodfaith Edit Models
Open, Needs TriagePublic

Description

Project description

This work helps to increase the distribution of the Recent Activity module which is part of the New Dashboard for Moderators. It aims to deploy the revert risk model and the RecentChanges filters powered by it to wikis which don't have the damaging/goodfaith Edit Models T348298#11202425

Project-level Hypothesis

If we deploy the Revert Risk filter to +150 additional Wikipedias that currently lack damaging/goodfaith models, then we will see an increase in moderator patrol counts for contributors who use the personal moderator dashboard compared to those who don't get access to the dashboard.

In order to release this we need to:

  • Create a list of wikis that don't have damaging/good faith edits models and decide the ones to deploy to - All of these P84306, including the ones that have FlaggedRevs installed T402780#11392631
  • Identify if there are any changes to the existing deployment process which was started in T348298 - This should remain largely unchanged and the same process will be used.
  • Machine-Learning-Team will run analysis on selected wikis to get thresholds and add these to MediaWiki config. T408607
  • Add the thresholds to each wiki(We're going to start with thwiki and then roll out to other wikis.) T409438
  • Turn on the models at specified wikis and run the scripts to backfill edit scores. These have been batched to make the rollout more manageable.
  • Schedule the deployments to the identified of wikis (this can be done in batches if need be)
  • Test that the filters are working.

Related Objects

Event Timeline

Restricted Application added a subscriber: Aklapper. · View Herald Transcript
DMburugu updated the task description. (Show Details)

Create a list of wikis that don't have damaging/good faith edits models

I grabbed this data a few weeks ago, uploaded here: P84306

Thanks for that. What are we using as our cut off for Wikis to consider for the rollout? Are we looking at the monthly active editors or monthly nonbot edits columns?

Will you be using the same definition for the filters as described in T392148: Run analysis to retrieve thresholds for high impact wikis to deploy recent changes revert risk language agnostic filters to? In that work a single filter was defined, and its threshold was configured to ensure fewer than 15% false positives.

I see there are wikis like Egyptian Arabic which has less than 30 active editors but over 50,000 edits

wiki namemonthly active editorsmonthly nonbot editsdatabase code
Egyptian Arabic Wikipedia26109,376arzwiki

Compared to Thai Wikipedia which has 10 times more moderators but less than half of the number of edits Egyptian Arabic has.

wiki namemonthly active editorsmonthly nonbot editsdatabase code
Thai Wikipedia35442,200thwiki

@isarantopoulos Yes, we'll be using the same false positive acceptance rate. We haven't discussed changing it yet.

DMburugu updated the task description. (Show Details)
DMburugu updated the task description. (Show Details)