2019 talk:Research/Discovering and Analyzing Wikimedia Images

From Wikimania

One tool that I would find really useful is a "find other photos of this same object" tool. Lots of uncategorised images on commons turn out to be of things that we already have images of, especially unidentified vehicles and sometimes buildings. Another tool, from the wikipedia end, or other user sites would be "find better images" so a little pop up would appear when browsing Wikipedia saying when an image could be upgraded with a better one from commons. At its simplest this could work on categories and clasification, so if you have a featured image with the same categorisation as the image you are looking at there is a reasonable chance that the featured image would be an improvement for that article. Of course if the article is fog and the image of a building in the fog you wouldn't want to replace that photo with one taken on a clear day. WereSpielChequers (talk) 16:52, 5 August 2019 (UTC)[reply]

Thanks @WereSpielChequers:! these sound like very interesting projects to start exploring during the Hackathon -- If you like, I can propose them as mini-projects for the Research table, and I am happy to supervise and help out with the implementation. Miriam (WMF) (talk)
Thanks Miriam, that would be cool, unfortunately I won't be at Wikimania this year. Obviously both tools would need to be opt in, but in wikipedia terms, upgrading images should be an uncontentious route in for newbies. A third tool would be to use geocodes to subdivide the completely uncategorised images into uncategorised by country, US state etc. A pop up for location based on geocode would also be helpful, when I'm categorising I spend a lot of time using various maps to find places by geocode. WereSpielChequers (talk) 11:35, 6 August 2019 (UTC)[reply]
Hi @WereSpielChequers:, thanks for your feedback again! And sorry to hear you can't make it to Wikimania. Following your first suggestion, I modified the presentation a bit, so that now we have a section on how to compare and cluster images using visual content. You can check out the slides here: slides. I also put some code online. Hopefully this can help people working on some of the tools you have discussed, even without attending the tutorial. Miriam (WMF) (talk)