Page MenuHomePhabricator

How many edits would Automoderator revert per day at different caution levels?
Closed, ResolvedPublic

Description

As part of our model testing process I think it would be helpful to understand how many edits we can expect Automoderator to revert. This will be useful for the community to understand Automoderator's potential impact, and we can compare with anti-vandalism bots, which we researched in T341857.

This should incorporate the internal configurations we already want to consider, i.e. not taking action on:

  • Administrators
  • Bots
  • Self-reverts
  • New page creations

Question:

  • What would the average number of daily reverts be for Automoderator ...
    • At the different caution levels we have in our test (0.97, 0.975, 0.98, 0.985, 0.99)
    • On the English and Indonesian Wikipedias

Event Timeline

KCVelaga_WMF changed the task status from Open to In Progress.Oct 16 2023, 11:12 AM

@Samwalton9-WMF

Average daily number of edits Automoderator would potentially revert per day at different thresholds (based on 2022 data)

wikit-0.99t-0.985t-0.98t-0.975t-0.97
English Wikipedia46100165235308
Indonesian Wikipedia12468

Notebook

Thanks for this @KCVelaga_WMF! Could you repeat the analysis for the German and Japanese Wikipedias? These communities are also engaged with Automoderator so we decided to proactively provide data/analysis for them too. It would also provide us with a couple of wikis between these wiki sizes :)

@Samwalton9-WMF Sure, here is the info for all four wikis

Wikipediat-0.99t-0.985t-0.98t-0.975t-0.97
German2471013
English46100165235308
Indonesian12468
Japanese1518212631

@Samwalton9-WMF Sure, here is the info for all four wikis

Wikipediat-0.99t-0.985t-0.98t-0.975t-0.97
German2471013
English46100165235308
Indonesian12468
Japanese1518212631

Huh - I'm surprised to see de.wiki with such low numbers here. I wonder why that is. FlaggedRevs maybe?

Huh - I'm surprised to see de.wiki with such low numbers here. I wonder why that is. FlaggedRevs maybe?

I think all these numbers are relatively low, ~1 % of all reverted edits does not seem like much help. As I've commented here, it might be because of heavy use of Abuse Filter.
Efficient and timely blocking also helps. On hrwiki we've had an average of 200 (IP) blocks a month, since May 2022 they're mostly done by Abuse Filter, which lead to a huge reduction of vandalisms overall.

mpopov subscribed.

@Samwalton9-WMF: It looks like @KCVelaga_WMF completed the request so I am resolving this task. If this is incorrect, please re-open and let KC know what is missing. Otherwise if you have follow-up/additional requests expanding on this one, please create a new task and check with KC about its priority relative to existing planned work.