Some of you may be aware that I've only recently "come back" after having left two years ago. After I departed DA staff, I moved onto other ventures and divorced myself from the internet in general. Now I've come back and found that in two years' time, not a great deal has changed. As someone in IT and high technology, this is pretty astonishing to me. In an era where folks are already looking at a sort of "Web 2.5", deviantART is still working on "Web 1.7". (Yes, I pretty much just made that up right there.)

My problem is that the success of deviantART as a business is symbiotically linked to the success of participating artists. Right now, everything is a big stew and the flavors don't all mix. In particular, I'm less than thrilled about the Collections feature because I don't see how this translates into dollars and cents for fine- and pro-artists. It's boondoggle for the community, and that's important. But DA is losing ground on the internet as the place to buy art.

I've had a couple conversations with people recently, and thus far the people I've been talking with have yet to turn the gun on me and ask: "Okay, you clearly feel strongly about this; what would you suggest to actually make things better." It would be a fair question, and you'd be wrong about me in more ways than you know if you think I haven't got an answer.

The answer is clear: mathematics. Yeah, that subject most of you were all too eager to discard after high school.

Let's back up a step and talk about art preference. If you look around my favorites, they're fairly diverse. I like to pick and choose, but what you won't see is a lot of fan art or emoticons. These things have their place in an art community and that place is, in fact, important. I'm just not a part of that scene.

So, you look at a piece of art. In this case, I'm looking at the one below. Broken Dawn by `arcipello.


If you find yourself looking at this piece of art and loving it, wouldn't it be reasonable and even valuable for you to find things that are similar? Damn skippy it would? But how do you get from A to B? What do you use to determine what to recommend to the viewer that they may also like?

Note that I used a key word there: recommend. This problem has been solved before. You build a recommendation engine! How do those work? They tend use a technique called the k-nearest-neighbor algorithm. This is an algorithm that lets you compute, say, that someone who purchased one item also purchased this item. Another example would be to say that someone who reviewed this item positively also reviewed this one positively. You've seen this in action before if you've ever visited a major etailer like Amazon or a movie site like Netflix.

Why not use it here? If you liked Broken Dawn, why not have an "Exploring deviantART" panel that would calculate there is a strong correlation to like exit scene by `spyroteknik (below) as well?


What's the end effect? Well, if you started with Broken Dawn, you could skip from deviation to deviation, following a common theme. What's that theme? People who favorited what you're looking at also gravitated towards the recommended pieces of art. So, you would skip from `arcipello to `spyroteknik to ~ANTIFAN-REAL to `nighty to `tigaer. I don't enjoy fan art; it's not my thing. With a k-nearest-neighbor algorithm, I can explore DA without running into much of it.

Or let's say fan art is your thing. You'll find yourself "exploring" around recommendations that reflect that. Into high tech skins? You'll see more of those. Into off-beat urbane art like `jasinski's? You'll find your way to ~melocopter and friends in no time.

There's real value in this. A lot of the artists I named sell prints. If you're looking to deck out an empty-walled room with DA prints, it can be frustrating to try to find things that work well together. DA Prints Store just doesn't do enough to help you find common pieces of art. Sometimes, you find one artist you want the room to be a shrine to. Sometimes, you want a mixture of artists. You've got to hunt, to cherry pick. Recommendations are manual and don't correlate to a specific piece.

Here are some exmaples of the kind of things you could find yourself hopping around to with a good k-nearest-neighbors algorithm. It all depends on how well its implemented and how strongly correlated the data actually is. These are pulled from my favorites to exemplify the similarity principle.



DA engineers, you have the data set to do this. You have the community tools used to power this. You have the ability to let users of DA drive their experience of actual art on DA. All you've got to do is do it. I've been gone two years and came back to Collections, a version 2.0 of the Favorites system. Come on, guys. Bring DA up to speed.

Thanks for your attention. And now, back to work. If I said any more on this subject, I'd damn well have to charge for it.