Compare app-uploaded and mobile website-uploaded pictures

[nicolas-raoul]

In order to understand (or demonstrate) how useful the Commons Android app is, it would be good to compare pictures uploaded:

• Via the Commons Android app, meaning they have their creation history has the Android app edit tag (example).
• Via https://commons.m.wikimedia.org (presumably from a mobile device), meaning their creation history has the Mobile web edit tag (example).

Things that can theoretically be algorithmically compared:

• Number of pictures
• Proportion of pictures from under-represented regions (among pictures that have metadata with which we know the region, for instance with coordinates or categories hierarchy)
• Proportion of pictures with categories
• Proportion of pictures with depictions
• Proportion of pictures with coordinates
• Proportion of pictures with a caption
• Proportion of pictures with a description

Things that probably require human judgement:

• Title quality
• Categories quality
• Picture quality

[misaochan]

This data would be interesting indeed! This would require the usage of a bot (with the corresponding bot privileges) I guess?

[nicolas-raoul]

Hopefully a part of it can be done by some query similar to https://quarry.wmcloud.org/query/10587

For the rest, rather than use a crawling bot (very costly in terms of server resources) we should probably download Commons metadata and parse it locally. It should be one or several of the files at https://dumps.wikimedia.org/commonswiki/latest/ , not sure which one(s). Not need for any privilege, not even a Wikimedia account.