Finding Great Domain Names with Mechanical Turk

(Cross-posted from Domain Pigeon)

Overview

  • Finding great available domain names just got a little bit easier with some help from Amazon’s Mechanical Turk community.
  • The community rated the pronounceability of Domain Pigeon’s entire collection of five and six letter .com domain names.
  • It worked well: Check out the six letter domain names sorted by pronounceability
  • Five is better than six! Sign up for an account to explore all the five letter domain names.
  • Added bonus: To celebrate, we’re adding 1,000 new five letter domain names per day from Monday, May 11 through Friday, May 15, 2009 at 3pm EST.

Background

Anyone that’s attempted to search for an available domain name recently can attest to how hard it is to find a good name. If you’ve ever spent any time on Ajaxwhois.com, you know what I mean. You can futilely sit there for hours typing various combinations of words until you finally get so frustrated that you settle on something that’s long, hyphenated, unpronounceable, and has six to ten numbers tacked on the end. Not exactly your ideal name.

Domain Pigeon was created to help alleviate the pain of finding a good domain name. We find available domain names and then list them on this website so that you can browse them and find ones that appeal to you.

To be clear: it’s not difficult to find available domain names, but it is hard to find good ones. Smack your keyboard against your desk about ten times until you have something that resembles xifs-duodis9.com and chances are you’ll be able to register it. But that’s not what you want. You want something marketable. Something remarkable.

We’ve used various methods to find quality domain names. For example, one way to come up with domain names is to simply add “i” or “e” to the beginning of a word and check whether that domain is available: iPhone.com, ePhone.com, iStore.com, eStore.com, etc. We’ve got plenty of those (‘eSlanting.com,’ anyone?), but that’s not what most people want. Another method is to check every group of five letters within a word, hoping that it produces a decent name. For example, using the word “alphabetic” we could check the availability of alpha.com, lphab.com, phabe.com, habet.com, abeti.com, and betic.com.

When you do that though, you get a lot of garabge like lphab.com. It’s barely pronounceable and when added with the other names it adds a lot of clutter to the site. The problem is that its not that easy for the computer to tell which names are good and which are bad.

Until recently, I’ve simply let the visitors sort it out. I posted everything using the philosophy: Don’t like it? Don’t register it.

A Better Way


I recently saw Mike Culver, an Amazon Web Services evangelist, speak at the Philly Emerging Tech conference. I had heard of Mechanical Turk and knew the concept, but was curious to learn more so I attended his talk on it.

For the unenlightened, Mechanical Turk is, in Amazon’s words, “a marketplace for work that requires human intelligence.” Basically, its a system that enables thousands of people around the world to help you with tasks that computers have a hard time doing. In exchange for their help, you pay them some small amount. For example, its very difficult for computers to analyze the contents of a picture so you could use Mechanical Turk to tag images or to identify pornography (yes: someone will pay you to look at porn).

On the drive home that day I was thinking about ways to use it for Domain Pigeon. Then it hit me: Why not have the workers (the people who complete the tasks) help identify good domain names?

I decided that above all else, pronounceability and length were the two most important characteristics of a good domain name. If you have to spell out any portion of your domain name when telling it to somebody, it’s probably not a good name. Additionally, you want it to be as short as possible. After all, “thisisanawesomedomainnamebutyoudprobablyneverwantit.com” is easy to pronounce, but it’s hard to remember and even harder to type.

So, I went about creating a task (a Human Intelligence Task or HIT in Turk-speak) that would ask people how pronounceable thousands of different words were. The words were made up of domain names that I have already posted or plan to post on Domain Pigeon. I limited the domains to ones that were five or six letters long because alas, there are no available .com domain names less than five letters long.

Here’s a brief summary of what this entailed:

1. Create a template for the HIT:

After a worker completes a task, the person who created the task rates the workers’ performance. The better the workers do the higher their HIT approval rating will be. When you create a task, you can specify a minimum HIT approval rating which will determine the quality of the worker that can work on your task. Generally you want to keep it pretty high so you can reduce the number of garbage answers, but remember that the higher you make it, the less people who will be able to work on your task and the longer it will take to get completed. I lowered it to 0.80 to allow more participants.

As you’ll see in a second, I added 40 domain names per HIT. That means that when someone accepts one of these tasks, they have to rate the pronounceability of 40 words. I set a reward of 4 cents for this grueling task and I also said I wanted each HIT (each group of 40) to be rated 4 times. I wanted several people to rate each domain so I could then average the scores to get a more reliable rating.

2. Design the layout:

You can use their editor to create a layout, or if you’re familiar with HTML you can create your own. A basic knowledge of CSS is also helpful because, well, you should be nice to the workers and it doesn’t take much work to add a little style.

The ${domain1} format is a variable which will be replaced when I upload the data…

3. Upload the data:

When you’re happy with the price and the layout, it’s time to upload the data you want it to use.

For this task I wrote a few lines of code that would pull all of the available five and six letter domain names from my local database and output them as a Comma Separated Value (CSV) file where each row corresponds to the data used in a single HIT.

For this task, there were 40 columns and 590 rows for a total of 23,600 domains.

4. Publish and wait

This is what I see:

This is what the workers see:

5. Export results, analyze

When you’re done you can export the results, which you can them import into Excel and analyze to your heart’s (or rather to your ability’s) content.

Here’s what the analysis looked like for this this task:

The Result

You can now sort five and six letter domain names on Domain Pigeon by pronounceability. While not perfect, it does make it a lot easier to find the cream of the crop domain names on the site.

Guests: Check out the six letter domain names sorted by pronounceability
Members: Also check out the five letter domain names (Not a member? Sign up)

To celebrate the addition of this new feature to Domain Pigeon, we’re adding 1,000 five letter domain names per day for the entire week of 11 May, 2009 through 15 May, 2009.

Enjoy the domain names and happy hunting!

Domain Pigeon Adwords

Your first time is always special:

I set a monthly budget of $250 and entered about 15 phrases I’d like Domain Pigeon to pop up for.

I do realize that it’s currently the third result for this search term and that paying for advertising on this page is probably unnecessary, but what the hell, let’s see what happens.

Finally: Credit Card Processing

On February 8 I wrote the following:

Before I start adding more features I’m going to add credit card processing to the site. The percentage of people that follow through to Paypal to complete their registration is abysmal.

Well, I was lazy and abandoned it in favor of adding some shiny new feature.

After the CNet traffic two weeks ago I decided to take another look at the conversion rates.

To make a long story short: only 10% of the people were completing their purchases. Ouch. I think Paypal is difficult to use for people without Paypal accounts and I think most of them didn’t complete their purchases. There could have been usability problems on my end too, but I think it’s minimal in comparison.

I said enough is enough, I’m not doing anything else until I get this done. So, I spent the last two weeks adding credit card processing to the site. Hallelujah.

I have to thank Ryan Bates from Railscasts for his excellent videos on Active Merchant.  Without his tutorials this would have taken much, much longer.

Here’s what it looks like:

I’d also be remiss not to thank 37Signals, for it was the registration page for Backpack that was my inspriration for this design.

I’m thrilled that its done and that I can get back to adding new features.

P.S. This is a good read: The 5 Things I’d Tell My 21 Year Old Entrepreneurial Self.

TechRadar Article

From TechRadar:

Twitter – in case you hadn’t noticed – is becoming so mainstream there’s now even at least one third-party business dedicated to finding the best available account names.

The Twitter open account search comes courtesy of Domain Pigeon, which previously focused on searching for unregistered URLs, so extending that to the microblogging phenom is hardly a stretch.

Three letters

Naturally, the most sought-after Twitter usernames are the shortest ones – three characters is the minimum, of which about 18,000 are still available, we’re told.

Before you rush off to see if xxx or abc are still up for grabs, be warned that Domain Pigeon requires a one-off membership fee of $40 (£27).

Considering that, maybe your next Tweet can be something like “@domainpigeon You’ve got to be kidding.”

Not kidding. :D