Jump to content

Topic on Talk:Growth/Personalized first day/Structured tasks/Add an image

More ways to identify images & alternative approach

2
Prototyperspective (talkcontribs)

Great project! This is really interesting but I think many ways and the best ways to identify missing and fitting images are still missing. Could you please clarify whether these are included and if not, as probably the case, please add these?:

  • Media set on the Wikidata item should in most cases also be in the Wikipedia article. Some Wikipedias seem to have some templates that automatically use whatever image and video is set there. (example of item with barely used useful image)
  • There are many articles with no or only few media but good-quality media on WMC. This doesn't apply only to images but also to videos and it would be good if this was extended to also include videos (as in "Add a media file"). Also the current method seems to only work when some Wikipedia already uses the media file (?) but not when all language versions don't have a media file.
    • It could start with media located in the Wikimedia Commons category of the file or a subcategory of it that have indicators for being of high-quality and illustrative of the subject such as being a Featured picture or being a media of the day (see more like that here) or being a chart in a category like Charts showing data through 2022 (recent year) and so on (this could be extended and improved upon over time)
    • For example, Hakkaisan Ropeway has no media but its WMC category has two files of which the video was a MOTD (and including English captions). This hasn't changed after the file has been motd and this is normal, there are many examples like it.
  • If the image is already used in another Wikipedia (especially EN WP), please suggest that text as machine-translated caption which the user can adjust. For videos that have been MOTD one could use the caption used there and one could also make use of the media captions and description on WMC.
  • It could also show many images from the corresponding WMC cat and ask the user if any and which of these is(/are?) useful for illustration. Note that many articles even lack the Commons category template and that should be added as well if it's missing so the readers can more easily find (more) media if they see/found this link which they rarely do.

However, I think the approach isn't efficient and scalable. There are over 300 language Wikipedias, imagine how much time it would take to add just one image to all of them. This approach wastes a lot of time and could more often lead to low-quality or inappropriate additions than with other ways. It may still be useful until something better is implemented and as something complementary to it but I think a better way would be to build ways by which media are added or changed for all language versions at once / centrally. The most straightforward way for this would be to display the image(s) and video(s) set on the corresponding Wikidata item on the articles if not included otherwise. It could be shown below the lead and users could then remove or move these files. My proposal for this is here: Community Wishlist – Including media files set on Wikidata item in Wikipedia articles across languages by default.

KStoller-WMF (talkcontribs)

@Prototyperspective Thanks for the feedback and questions! I'll try my best to respond to each of your points:

Media set on the Wikidata item should in most cases also be in the Wikipedia article. Some Wikipedias seem to have some templates that automatically use whatever image and video is set there.

We could certainly consider something like that, although the main purpose of the Growth team's "Add an image" task is actually about giving newcomers a structured and easy way to get started editing. So although it's not as instantaneous as the option you are suggesting, it's helpful to have a task pool of good suggestions for newcomers. You can read more about how the task helps increase newcomer participation: Add an image/Experiment analysis, March 2024.

...it would be good if this was extended to also include videos

Agreed. This would take additional work, but it's an improvement we can consider in the future.

Also the current method seems to only work when some Wikipedia already uses the media file...

This is one source of the algorithm, i.e., Wikipedia article lead images. Other sources are detailed here.

It could start with media located in the Wikimedia Commons category

The image suggestions algorithm actually does source from Commons. From both the Commons category (P373) (from Wikidata) and from Structured Data on Commons, namely Depicts statements.

please suggest that text as machine-translated caption which the user can adjust.

This is something we've considered. It would take more work, but I agree it's an idea worth exploring. (Although I have heard from some people who are concerned with this approach since they worry newcomers will just accept Machine Translation suggestions even when the translation is awkward).

There are over 300 language Wikipedias, imagine how much time it would take to add just one image to all of them. This approach wastes a lot of time and could more often lead to low-quality or inappropriate additions than with other ways.

We use a confidence score to help filter out less relevant suggestions, see here.

The Growth team's feature is certainly partially about helping illustrate articles, but the underlying purpose of the task is to help newcomers get started editing. By adding images in this way, the image is reviewed both by the new editor, and by a patroller checking the newcomer's edit. It's certainly not perfect, since new editors are learning and sometimes make mistakes, but it helps ensure there is some human review before images are added.

Thanks for all of the feedback and questions! The Growth team isn't actively working on this task now, we are hoping that we can start to scale this feature to more wikis soon, and I definitely agree there are many opportunities to continue to improve and expand the feature!

Reply to "More ways to identify images & alternative approach"