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.
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.
Comments
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.
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.
Or just take the median.