Re:Show HN: 51,000 6 character domains
by jqueryin on Friday July 06, 2012 @10:59AM
(#4207999)
The 5 letter space is a bit more crowded, but I could definitely follow up with that next. I might try running combinations of bigrams and trigrams.
[ Parent
| on HN ]
It's a fun exercise. I've tried a lot of similar stuff before (http://www.kevinohashi.com/17/04/2011/other-24000-available-...). The biggest problem I ran into with n-grams was trying to evaluate what was good in an automated way. Using stuff like Norvig's data to try and weight became futile. Stuff like ededed.com would rank exceptionally high if you weight by frequency. That's a bad name. I tried not counting dupes, giving dupes decreasing value as they increased, etc. Nothing really produced results that weighted nicely when I looked at them. This strategy did find the crap names at the bottom though. The problem was the 75% or so that was left was mixed.
I am curious if you tried to algorithmically sort the lists by any sort of nice/value/brand factor?
It'd be a fun exercise to weight the domains. I've personally considered (but haven't gone through with) trying to implement a metaphone or soundex algorithm to determine if the resulting words "sounded" English based on comparison with dictionary data phonemes. Phonemes in general are the reason I decided to go with combinations of high frequency digrams and trigrams.