Friday, February 03, 2006

Content 2.5: Farmer's Market

Who Are You? What Do You Want?

I was going to approach this slightly differently, but Raph made a post about social feedback, so I decided to rearrange my chapters a bit. This isn't a full chapter: it's a comment on a specific kind of content distribution system. It is assuming a game thick in player generated content. :)

When it comes to distributing good content and limiting bad content, people generally want a ratings system. They want a ratings system that is zero-maintenance. It should work automatically. Which means, of course, that it runs on player juice.

Unfortunately, player juice tastes like the player. Some players are tasty apples, some are tart lemons, and some taste like shit. The problem is that a computer can't really tell who is who.

Oh, there are systems that can try, but they all have holes and are subject to farming. Most such systems rely on the players never really figuring out how the rating is calculated - which is an awfully poor thing to rely on, as any Linux fan will tell you.

So you get farmers who run around totally hosing the balance of the system. To correct for griefers, you put your emphasis on positive feedback, but this solves nothing, because farmers will still come out on top.

You can use personalized systems, such as only seeing the ratings made by people you've rated. But, as Raph points out, new users don't have personalized systems. You can get around this, but only if your whole game is based around it - not what most people want.

So is there some way of responding to content creation in a way which reliably drags the best into the limelight and relegates griefers to the dark, dank corners of the world?

Actually, yes. There are several. Because we don't care about a one-axis rating system, like most people think in terms of. One-axis rating systems are used to support a full-depth environment.

For example, if you go to eBay, you're looking for something specific. Like, say, a dart board with your favorite politician's face on it. The system finds the dart board for you, and then you can look at all the offers. Here's one that costs a bit more, here's one that costs a bit less. That seller has a high rating, that one not so much.

In this case, the rating is what I call a single action rating. It's intended to give you an idea of how that seller will act on that one act: the act of sending you the things you paid for.

Providing content is not a single act. It is a deeply flawed idea to assume you can simply rate content on a numeric scale. I hate racing games, but that doesn't mean racing games are bad. Similarly, I'm a sucker for architecture, even if it's not really very good.

eBay doesn't just pop up with a long list of random crap with a numeric rating. The majority of eBay's software is for far more complex search algorithms. Similarly, your rating system shouldn't just pop up with a number.

Of course, your game is fundamentally different from eBay. Most people who go to eBay know exactly what they want and consider it an important transaction. Most people in your game won't know what they want and won't consider it an important transaction.

This lets you use extremely vague "search" algorithms. It lets you use "geographic" rating systems.

I mentioned these in earlier posts. A geographic system is one in which content is created locally, then slowly distributed as it is carried around by its popularity. For example, if you make a new kind of nifty space ship, at first it's only you using it. However, others want to buy it, and so you make more. Soon, there's dozens or hundreds of players flying these ships all over the place. A newb looking for a good ship will naturally see this option.

On the other hand, if you create a ship that looks like a giant flying penis, it's unlikely to spread very far. First, because few players will want to buy it. Second, because many players will attack and destroy it out of disgust.

Obviously, this depends on what kind of content you allow people to create and what kind of barriers you have in place. In many games, making a ship that looks like that would be simply impossible. However, the idea is the same:

Griefer content can't get very "far" in your "terrain" because it can only be spread by willing hosts.

This has the problem that a large group of griefers can spread their poison throughout the land. However, because of the level of coordination that is required, it is relatively easy to catch and annihilate these teams with a moderator. Certainly easier than having dozens of moderators rating each new piece of content.

This, of course, doesn't stop griefers from running around killing people. But it does keep them from spreading nasty content, to a large extent. If you go looking, you'll still find it. But if you stick to well-traveled paths, you're fine.

Well-traveled paths are, in fact, the major issue here. A ratings system is simply an attempt to point out well-traveled paths. And any geographical approach requires paths. How can you do this?

The basic idea is that a player will suddenly find himself on a network of roads leading to all the cool content. There's two pieces to that: creating the roads and getting the players to them.

One obvious solution would be to have players have an "inventory" of content (a public and a private inventory). Any other player looking at them can see their public inventory, along with how many other players have it in their inventory (public or private). This will introduce them to the content (they can then go find it) and will also give them other ratings by the same person. Meaning, if they liked the first rated one, they're likely to like the second rated one as well.

This is similar to a social network system, but it lets the new players tap straight into an existing social network without disrupting it.

What this does is simulate conversation between the new player and the old player. Functionally, it is letting every new player say, "hey, you, experienced guy: what's fun to do around here?" and guaranteeing that the experienced guy actually replies with something useful. All without disrupting the experienced player.

Disrupting the experienced players is what kills most attempts to do something like this, because few experienced players actually like teaching a newb, let alone dozens of them. (I wrote an essay about this a long time ago, actually.)

By isolating the most common and important questions a new player might ask and letting him ask them without bothering anyone, you allow the new player to merge into the framework of the game with ease. It used to be this was stuff like, "how do I cast 'Magic Missile'?" and "how do I get to Chunya Cave?" Thus instruction manuals and maps/waypoints were born.

With the added complexity of player generated content, however, the questions are much more personal. Not, "what content is the best?", but "what content that is in the genres I like is the best?"

By allowing more experienced players to "save" their "potential conversations", then letting new players access them without bothering the older players, you've created an inviting environment for both experienced and new players.

This can be done for lots of different kinds of content, and in lots of different kinds of ways. It's the heart of a geographic distribution model. It can still be kinda-sorta farmed, but the tactics are vastly more acceptable and less dangerous.



Eric Poulton said...

A good example of this is DeviantArt. I don't think there's a bigger collection of terrible content in the world, but they've managed to offer some fairly effective ways to find the good stuff.

The Favourites system they have works well because people tend to pick Favourites that have the same flavour as their own art, but are at least a bit better. Even starting at a terrible artist, by clicking on one of their favourites you'll usually be able to find a slightly better artist, and so on until you'll usually find some very good work you haven't seen before.

DeviantArt has hosts of other problems, but the Favourites system tends to work well.

Craig Perko said...

That is a perfect example. :)