Reading/Multimedia/File feedback

This is just an idea collection page from 2013 at this moment. This is based on a discussion between Robla, Fabrice and Sage Ross.

The issue

When browsing a category, images are shown in alphabetical order. Usually when browsing a category, you want to see the "best" images first, as they are the one's you are most likely to use. A significant portion of people coming to commons want to find a high quality image of some topic. We should help them by showing them the "better" images first.

Furthermore, there is no complete rating scheme of media on commons. Really good pictures (Quality images, Valued images, featured images, etc) do get identified, but that is only the cream of the crop. We have no information on how good an arbitrary file is.

Potential Solution

Have a rating system targeted at visitors (Not primarily targeted at contributors). This would not aim to be a venue for provided constructive feedback. It would simply be a way of gathering information on if one file is better then another so we could mechanically sort them.

Obviously what is a "good" image is up to much debate, and not all useful images are necessarily "good". The goal would be to get an approximate idea of how well an image compares to other images.

Once we have ratings for some files, we can have a (perhaps) optional mode in categories to sort by rating. We perhaps could even do other sorting methods like sort by how many times this file is used (downside: The extra writes to the categorylinks table could have performance issues), file size, etc. We could also have special:BestRatedFiles and Special:WorstRatedFiles.

Solution A (thumb up/thumb down)

Have a "thumbs up"/"thumbs down" box on file pages

From there sort things using the method described at http://www.evanmiller.org/how-not-to-sort-by-average-rating.html

• Good:
  • We'd get feedback on well used files much quicker, and would probably get a usable amount of data fast. Compared to the compare two random approaches where the ratings would be spread out amongst 17 million files
• Bad
  • Might turn into a popularity contest (since people would only see the rate button on images that are actually used, so we might identify used images instead of surfacing interesting but currently unknown images).
    • If it did turn into a popularity contest, we'd probably be better off directly using popularity data. (global image links, or even better number of hits if we could swing it)
  • Unclear what sort of criteria we want to rate it on. "Image is good" - what is a good image, "Image is useful" - useful to whom. Useful how. Would an anon reasonably be able to tell if an image is useful. "Useful to wikipedia" - the world isn't just wikipedia, but even still its hard to determine if something is useful to wikipedia. "Do I think the image is pretty" - realistically this is what people are most likely to vote on regardless of what we tell them.
    • It seems like it would be much easier to get consistent results if we told people to compare images, instead of asking them if an image meets some absolute standard.

Solution B

Instead of getting users to rate files in isolation, have them compare files.

• Would solve problems of different people having different standards, since we're comparing in relative terms, not absolute terms
• Could have a special page Special:Which_is_better? that shows two images chosen at random (or from same category?) and asks users which is better
  • Good: It could be a fun way to explore commons, discover new images, etc. Could even be almost a game.
  • Downside, users could be asked to compare apples and oranges (or which is better - An audio recording of an opera or a picture of a firearm).
  • If images chosen at random, could potentially have offensive selections (which of these sexually explicit images is better? Which of these religious figures is better? etc)
  • If done for random images out of the pot of 17 million images, may be a long time before we get useful enough information to do a ranking system. Doing specific categories might be a way around this.
• The idea would be to make it into a bit of a game that could be fun.
• Issue: How do we tally ratings for this. Would using the Lower bound of Wilson score confidence interval for a Bernoulli parameter method work if we're doing comparisons instead of good vs bad ratings? (Other useful links: w:Condorcet_method)

Demo of this

Potential problems

• Users might worry it distracts from serious work of commons
• Ratings might be considered not useful.
  • There are a lot of files, if images randomly selected