I found reading that new paper very interesting, and also disheartening.
It seems Humans are still much better at classifying things.  I am critical of many of the assumptions that were made, even by their Judge LLM constructed (which I fear was given sometimes wrong training).  As an example, some links were removed that actually provided value and made BETTER sense of our highly structured world and brains.

Case in point, they discuss "urban area" <--> "human settlement" or vice-versa.  It's clear to anyone that what defines "urban" to begin with is of extreme importance - and it differs often depending where you live in the world - so ranking has to be pulled into the picture to give some general consensus to a real world model (in all areas of the world, does "urban" mean inside a "settlement"... always?...everywhere?  We know the answer is no.)  But generally we have consensus that there are broad city areas, and human language terms that relate boundaries of those areas, and further terms that actually define those boundary areas.  Does "urban" itself focus on a boundary?  Sometimes, in many parts of the world, it does.  Sometimes not.  Regardless, there's consensus that "urban" would be inclusive of a "city" boundary, and "rural" would be outside a "city" boundary - most of the time 🙂.  So, exactly because of "most of the time" this implies that ranking should be applied to subclass relations.... instead of cutting and merging subclass relations, as the paper's WiKC model did.  What a shame.  It should have avoided cutting and merging, and instead liberally applied better ranking, as a first pass, in my opinion.

Still, BIG KUDOS, to the team, Yiwen Peng, Thomas Bonald, Mehwish Alam, et. al.  for even approaching such a difficult task and having to spend hundreds of hours in training and research combined - and attempting to untangle a mess of "instance of" with machines 🙂

- Thad


From: Mohammed Sadat Abdulai <mohammed.abdulai@wikimedia.de>
Sent: Tuesday, September 24, 2024 12:27 AM
To: Discussion list for the Wikidata project <wikidata@lists.wikimedia.org>
Subject: [Wikidata] Weekly Summary #646
 
Here's your quick overview of what has been happening around Wikidata in the
week leading up to 2024-09-23. Please help Translate. Missed the previous one?
See issue #645

Discussions

  • Open request for adminship: Andrei Stroe - RfP scheduled to end after 23 September 2024 12:41 (UTC).
  • New requests for permissions/Bot:
  • Other discussion: Enabling the CampaignEvents Extention on Wikidata - Proposal to enable the CampaignEvents extension on Wikidata to make it easier for organizers to manage community events and projects on the wikis.

Events

Press, articles, blog posts, videos

Tool of the week

Other Noteworthy Stuff

Newest properties and property proposals to review

You can comment on all open property proposals!

Did you know?

Development

  • Query Service:
    • We added a banner to the Query Service UI to inform people about the graph split.
    • We fixed an issue with visualisations caused by a library upgrade (phab:T375058)
  • mul language code: We continued working on the remaining issues uncovered in testing (phab:T373088)
  • Migrating to the Codex design system: We are in the process of migrating the Query Builder and Special:NewLexeme from Wikit to Codex.
  • REST API: Continue work on improving error handling and messages
  • Data access: We are preparing the survey for reusers to better understand which additional data access methods they need.

You can see all open tickets related to Wikidata here. If you want to help, you can also have a look at the tasks needing a volunteer.

Weekly Tasks

Read the full report

--
Mohammed S. Abdulai
Community Communications Manager, Wikidata

Wikimedia Deutschland e. V. | Tempelhofer Ufer 23-24 | 10963 Berlin
Phone: +49 (0) 30 577 116 2466
https://wikimedia.de

Grab a spot in my calendar for a chat: cal.com/masssly.

A lot is happening around Wikidata - Keep up to date! Current news and exciting stories about Wikimedia, Wikipedia and Free Knowledge in our newsletter (in German): Subscribe now.

Imagine a world in which every single human being can freely share in the sum of all knowledge. Help us to achieve our vision!
https://spenden.wikimedia.de

Wikimedia Deutschland — Gesellschaft zur Förderung Freien Wissens e. V. Eingetragen im Vereinsregister des Amtsgerichts Charlottenburg, VR 23855 B. Als gemeinnĂŒtzig anerkannt durch das Finanzamt fĂŒr Körperschaften I Berlin, Steuernummer 27/029/42207. GeschĂ€ftsfĂŒhrende VorstĂ€nde: Franziska Heine, Dr. Christian Humborg