Re:To Paul Graham
by sesqu on Wednesday June 02, 2010 @02:55PM
(#1398966)
How about instead of average comment score, you consider average likelihood of quality, treating the score (or its log) as odds? That is, w=avg(score/(1+score)). That way a single popular comment won't dominate.
[ Parent
| on HN ]
It appears he throws out the highest rated comment when computing the average. I assume that's because of what you're getting at: the distribution of comment scores tend to look like a power law, so the arithmetic mean of all the scores isn't really the right number to use.