Page MenuHomePhabricator

Analyse data on diff visits
Open, LowPublic

Description

Questions

  • How many unique users view namespace=0 diffs before one of them takes an action (Undo, Thank, Rollback)? What is the distribution of view counts?
  • How many unique users view each namespace=0 diff within 1 week of the edit being made?

For these analyses we probably only want to consider diffs made by humans, and probably only those made by new or unregistered users (the edits patrollers are primarily looking at).

Event Timeline

We may want to incorporate API requests in this analysis, to understand when diffs are viewed through patrolling tools like Huggle, through which a large portion of patrolling is done on some wikis. Probably worth splitting the data to understand where those views are coming from. Possible example query in T434795#12326581

We may want to incorporate API requests in this analysis, to understand when diffs are viewed through patrolling tools like Huggle, through which a large portion of patrolling is done on some wikis. Probably worth splitting the data to understand where those views are coming from. Possible example query in T434795#12326581

I started looking at this for Huggle. My not-a-data-expert analysis yielded 83% of edits by new editors (temp account or <=5 edits) on a given day (July 1st) being queried for by Huggle. That's surprisingly high! One reason it might be high is that Huggle might do that query in advance of actually showing a diff to a user (to generate the list of edits in the queue on the side), so the actual number would be lower. Not sure there's a good way to separate that out without having access to internal Huggle data.

I'm looking forward to the data, Sam. It sounds like it won't catch everything (e.g., hovering over the link with NAVPOPS), but it should help. Should it be limited to the mainspace?

Is there any chance of doing this also for newly created pages? (You don't really get "diffs" for the first version, so I don't think it would get picked up by the proposed change here.) I'd specifically like to know how many editors with the relevant user rights look at each article in [[Special:NewPagesFeed]].

I'm looking forward to the data, Sam. It sounds like it won't catch everything (e.g., hovering over the link with NAVPOPS), but it should help.

That's right. We could add some one-off analysis for tools and scripts like NAVPOPS but it might be hard to generate an exhaustive list and the analysis could be quite time consuming.

Should it be limited to the mainspace?

Yes, I think so - my sense is this is what is most pressing; reviewing edits in other namespaces isn't as important.

Is there any chance of doing this also for newly created pages? (You don't really get "diffs" for the first version, so I don't think it would get picked up by the proposed change here.) I'd specifically like to know how many editors with the relevant user rights look at each article in [[Special:NewPagesFeed]].

That would need to be a separate analysis - here we're recording data on the Special:Diff page specifically, so pageviews to an article would require separate instrumentation. Definitely an interesting idea though! Perhaps something that could be recorded in PageTriage as part of the code that loads the curation toolbar, so as to limit it to the relevant users.

I've created T438780: Track page views by admins and New Pages Patrollers via [[Special:NewPagesFeed]] for the idea about newly created pages. Let me know if you think that would be a suitable task for the Community Wishlist.

I think that could be a perfectly reasonable Wishlist item :)

I think communities would probably care more about / vote for features and bug fixes over data collection. But don't let my opinion stop you from creating a wish if you want :)