Hacker News Reader

Comment by user24 | Hacker News Reader

Be nice to merge it with a 2gram corpus sorted by frequency, so that word pairs that simply don't occur in English (like "autonomous frown", "voice glockenspiel", "harp speedboat") are less likely to be shown than words that might conceivably make sense.

Another idea might be to POS tag the words and only present verb noun combinations, but you'd have to play around a bit to see which parts of speech rendered the best domains.

Anyway, here are some nice sounding domains I've found with this tool:

historicalreading.com

brokencurrency.com

bagelmoon.com

sofacake.com

dancebrick.com

wholesalerreport.com

slimyslime.com (if you're a slime retailer, this is a perfect domain!)

and some not-so-good ones:

tramprugby.com (I would pay to see this)

aftershavejustice.com

beanslime.com

depressedtuesday.com (can we make this a national holiday?)

View on HN

I'm going to implement a sorting algo later (based on bayesian probabilities) and word popularity/top lists. But that's later..

The domains listed are only noun-noun or adjective-noun combinations. I might add verbs later.

2gram corpus is not that easy to implement, but here is my implementation with POS and frequency sorting. http://buzzdomain.com

Nice!

Small typo on home page: 'When you _clik_ the Search button'

IMHO 2-gram corpus is very easy to implement, what do you find difficult with it?

I've done this before. You're right. The CPU intensive part is the word search itself. Building the combinations and finding adjacent words, which you've already done. The ngram piece is straightforward by comparison. Just compare the domain (with any hyphens removed) against a dictionary of concatenated ngrams. You'll have to veryify that any hyphens occur on word breaks afterwards, but it's a small price to pay for the simplicity of a string compare (or key lookup). You will probably want to load all of your ngram dictionaries into memory, though, or the search will be horribly slow.

I have already done this, but it is difficult to squeeze the running algorithm memory. The current POS solution uses only 400MB memory. Also, 2grams are not that good as collocations.

That's interesting, because I think word pairs that simply don't occur in English could be a feature also. Maybe an on-off switch. I kind of like the domain voiceglockenspiel. Although several of your listed nice-sounding domains also seem to be pairs that simply don't occur in English as well?

> and some not-so-good ones:

> tramprugby.com (I would pay to see this)

Sounds like a good idea!

gradualunderpants.com was among the first I saw on the frontpage...

Reply on news.ycombinator.com