Page MenuHomePhabricator

Exclude articles that have article-level image suggestions from receiving section-level image suggestions
Closed, ResolvedPublic

Description

Description

As a newcomer completing suggested edits, I should never receive a section-level image suggestion if the article is unillustrated and has an article-level image suggestion.

Context

We only suggest article-level image suggestions for unillustrated articles. But we will generate and serve section-level image suggestions for both illustrated and unillustrated articles. It might mean that, if we send out an image suggestion for a section of an unillustrated article without or before a article-level image suggestion for the same article and a contributor adds it, the article will be flagged as illustrated during the next dataset update and thus excluded from article level image suggestion pipeline.

Acceptance Criteria
Given I'm a newcomer completing section-level image suggestions,
When articles are suggested,
Then I'm never suggested articles that already have a article-level image suggestion

Event Timeline

What would be easy to do on our side is to exclude articles which have a top-level image suggestion (ie. search for hasrecommendation:image -hasrecommendation:image_section instead of just hasrecommendation:image). Not sure how large a fraction that is of all unillustrated images.

What would be easy to do on our side is to exclude articles which have a top-level image suggestion (ie. search for hasrecommendation:image -hasrecommendation:image_section instead of just hasrecommendation:image). Not sure how large a fraction that is of all unillustrated images.

I might be misreading or thinking about this wrong, but I believe we want the inverse of the query you listed above: -hasrecommendation:image hasrecommendation:image_section.

Or in other words:

  • show me articles that don't have an "image" recommendation (so the article has an illustration): -hasrecommendation:image
  • do include articles that have section-level recommendations (hasrecommendation:image_section)

Looking at enwiki, that takes us from 104k to 102k articles. czwiki has a similar drop, from 11k to 10.5k.

As a newcomer completing suggested edits, I should never receive a section-level image suggestion if the article is unillustrated.

FWIW excluding articles with existing article-level suggestions doesn't guarantee that the article is currently unillustrated

@KStoller-WMF I wonder if we might move this task to a next iteration, so we have time to be more deliberate about how we choose to implement the business rules around it?

KStoller-WMF added a subscriber: Trizek.

Yes, in fact I think this is fine to shift to post MVP release.

And I'm open to considering an adjustment to the acceptance criteria if that vastly simplifies this task. Basically we should prioritize the article-level image suggestions since an image that illustrates the article as a whole is likely to be most impactful. But if we don't have any article-level image suggestions, perhaps adding a section-level image is OK? @Trizek let me know if you have any thoughts on idea.

Mmmh.

On one hand, it is better to have an image illustrating a detail than no image at all. If the image is super relevant to the section, but not to the overall article,

On the other hand, some people might want to see an image being added at the top of the article first, rather than at a section level. It is particularly true if the added image could fit the overall concept described in the article. Choosing where the image is located is not up to the user who adds it, but they would be blamed for not adding it at the right place.

I'd say: let's test it on our pilot wikis first and see what communities think of adding section images when no image is in the overall article. Based on their reactions, we would adjust our settings.

I might be misreading or thinking about this wrong, but I believe we want the inverse of the query you listed above: -hasrecommendation:image hasrecommendation:image_section.

Yes, my bad.

FWIW excluding articles with existing article-level suggestions doesn't guarantee that the article is currently unillustrated

We have 580K unillustrated in total on enwiki and 32K on cswiki (quarry; a bit narrower than the official definition which also allows e.g. infobox icons) vs. 88K top-level image recommendations on enwiki and 35K on cswiki. So it can be a big difference.

Change 924534 had a related patch set uploaded (by Gergő Tisza; author: Gergő Tisza):

[mediawiki/extensions/GrowthExperiments@master] Section images: Exclude articles which have a top-level recommendation

https://gerrit.wikimedia.org/r/924534

I guess if we are aiming for post-MVP with this, it's not top priority anymore.

Change 924534 merged by jenkins-bot:

[mediawiki/extensions/GrowthExperiments@master] Section images: Exclude articles which have a top-level recommendation

https://gerrit.wikimedia.org/r/924534

It only excludes articles which have a top-level image recommendation, not unillustrated articles which don't have one at the moment but are candidates for getting one.
Not sure if we want to do anything more; that would require help from Search.

KStoller-WMF renamed this task from Exclude unillustrated articles from section-level image suggestions to Exclude articles that have article-level image suggestions from receiving section-level image suggestions.Jul 14 2023, 4:33 PM
KStoller-WMF updated the task description. (Show Details)

It only excludes articles which have a top-level image recommendation, not unillustrated articles which don't have one at the moment but are candidates for getting one.
Not sure if we want to do anything more; that would require help from Search.

OK, thanks for the explanation! I updated this task to cover the portion of the problem fixed in https://gerrit.wikimedia.org/r/924534 and then logged a separate task to cover the remaining work (that would require help from Search): T341892: Exclude unillustrated articles from section-level image suggestions

I hope that sounds ok, and please feel free to adjust the title or description of T341892.

My perspective is that the work you did in this task covers the main issue we were trying to solve: making sure we aren't depleting the article-level image suggestions due to users completing section-level image suggestions. It might be ideal to also solve T341892, but I'm not sure if the effort is worth it. I won't plan on prioritizing this unless community members or someone on Growth, Search, or Structured Data thinks it's a priority.

Etonkovidova subscribed.

Checked in wmf.23 - the scope of the task is done; still, articles with many images are still recommended, but it's beyond the scope of this particular task.