CommonsDB gaining insights at Wikimania

This blog post is the fifth in the series about CommonsDB. If you don’t know about the project at all, I recommend checking out the first one, it also has a nice video introducing the project.

The pilot project CommonsDB, the registry with public domain and freely licensed images, is in the later stage, and it was with some excitement we wanted to test some ideas with the wider Wikimedia community.

The workshop

With this in mind, we’ve designed a workshop we titled “Putting CommonsDB into use“. To get everyone to the same knowledge level on how the system works, we had a brief introduction, and you can find these slides on Wikimedia Commons. From that, we broke into groups to have the participants try to solve different kinds of questions of importance to the CommonsDB registry and the Wikimedia Commons community. For example, to avoid that “bad” uploads to Wikimedia Commons are being declared to the CommonsDB registry, the groups discussed various factors that can act as signals to understand if an image is ready to be declared to the registry or if it should be held off for a while since the risk for deletion is considerable. Factors such as user rights, time since upload, metadata, and templates were discussed, and the groups came up with many suggestions that widened our view of what to consider. Taking these factors into consideration will be important if the project moves into a continuous phase with an aim to cover all current and future images on Wikimedia Commons. When criteria like this have been defined and tested for the CommonsDB project, they will likely also be useful for others that want to reuse Wikimedia Commons content in a streaming way and still want to safeguard their image selection somewhat.

Insights

The workshop gave us some good validation of our ideas and further insights to the wider community. One important outcome was that when it comes to trusting an upload, basically any extended rights were a good trust signal. Some other good trust signals that the groups came up with were watching for license templates being reviewed, if copyright information in EXIF data aligned with license templates, and if media was being reused in many places (invoking the many eyeballs principle). On the other hand, mismatches in EXIF data or files where the metadata itself is inconsistent or corrupt could be signals that these images still may need human review along with, of course, any deletion templates.

An idea for a backlog tool was also hatched. If we find matches in CommonsDB from another data provider, we can check their source to see if they have a file with higher resolution than the one on Commons and create a queue for a human reviewer to judge if the file we have should get a new version or if the file should be complemented with a new high-resolution upload.

The hackathon

While we at the previous hackathons had been hacking a bit ourselves to create prototypes on how finding similar images through CommonsDB can assist a user in the upload process, this time we instead were available for other developers looking to integrate such searches in their tools. There were two developers with related ideas. One who wants to create an interface for Wikimedia Commons focusing more on exploration and in that have an option to show similar images. While our previous use case mainly shows really similar images, thanks to the bit distance, it can be “loosened up” and experimentation with this can be done. Another developer was interested in doing similar matching as our prototypes but integrated into external upload tools rather than the upload wizard. Both of these would be very interesting, and we hope that the work continues after the hackathon.

CommonsDB contributing in other sessions

While at Wikimania, it’s almost impossible to not contribute in more places. Not only were the stickers we brought vanishingly popular, but at the Wikimedia Commons pre-conference we could also present the project and explain how image similarity techniques can be useful to the Commons community and GLAM institutions. For the impromptu Commons categorizers meetup, the introduction video was used as a walk-in-pause video, which set a joyful start to the meetup.