[Feature Request] average favorites

The total number of favorites mostly tell one thing about a (good) picture: how long ago it has been uploaded. If people order wallpapers by favorites (i.e. likes) they see the older pictures first, and if they happen to like them, that just aggregates more favorites, and the difference in favorites between newer and older pictures just gets bigger. (I still have the mental picture in front of me, where Barney Stinson, and some brain infograph were the top 2 pictures on wallbase for ever and ever.)

If you could introduce a new number that is the total number of favorites, divided by the number of days the picture has been on wallhaven, I think you'd get a much more impartial number. Even if there is a rush of favorites on recently uploaded pictures, which would place the new pictures on top of this list, this would normalize within days.

What do you think?

Even better there is the Bayesian average which uses global data to compute a realistic score [url]http://www.wikiwand.com/en/Bayesian_probability[/url] I couldn't write a pseudo code as it's a bit complex, there is a working implementation in Piwigo software [url]http://piwigo.org/dev/browser/trunk/include/functions_rate.inc.php?rev=30563#L129[/url] If course this need to be adapted with Big Data considerations :-)

The planned top list page will definitely do something smarter than just “most favorites” or “most views” (that's why it doesn't exist yet). The search engine we're using (ElasticSearch) is quite powerful and should allow for some pretty cool stuff.

Finding the ideal metric isn't easy, they all have their problems. For example following @1355970's suggestion and simply normalizing favs by age would mean that wallpapers that were uploaded a long time ago but only recently gained some attention would have almost no chance of making the top page (this is the approach reddit uses, btw). Also, wallpapers that already have a lot of faves may still stay at the top for an unreasonably long time.

This can be mitigated by only counting favs in a recent time range, like in the past X days / week / month. That will allow old wallpapers to have a new “bloom”, while wallpapers that already have all the favs they are ever going to get will make space for new “trending” ones. This approach isn't without drawbacks either: It can lead to the bandwagon effect, where basically once a wallpaper makes it to the top it gains so much exposure that it is constantly accumulating more favs and may stay at the top of the page for a long time. I think this is what happened at wallbase, although I'm not sure how their top page worked exactly.

In the end we have to find a metric that tries to take as many of these things into account while still being reasonably simple to calculate. May take some trial and error and tweaking around. ^^

Gandalf

I leave it to your capable hands. The only reason why I never learned any programming even though I'm a geek, is because I suck at math! :D

Anyway it's nice to know that you already planned something better than total favorites/views!

@1355970 said:

Gandalf I leave it to your capable hands. The only reason why I never learned any programming even though I'm a geek, is because I suck at math! :D Anyway it's nice to know that you already planned something better than total favorites/views!

You just need basic math for most programming. You can do just fine without advanced knowledge. It shouldn't stop you from learing it. Of course there are cases where a higher understanding would provide a better code/script. And sometimes it is indeed needed, but it's not like you can't learn some math as you go along. Besides, there are lots of tools, frameworks and scripts you can use to "cheat". It's not like you need to know algorithm in order to program basic PHP, SQL or JS.

Over to the problem. How about doing something like Deviatart, where you have "All time favourites" (sums all favs over all time) and "Popular/Trending" (sums favs based on days). If the wallpaper is older then the most amount of days allowed then it will only show up under "All time" or by searching.