The ideal measure would be number of upvotes of a comment divided by the number of views of the comment. That way you wouldn't get punished for making comments late in a thread or on a less well read post. Unfortunately measuring this statistic would require a lot of work to implement.
You could hook some javascript to the page's onscroll event. The javascript would send ajax calls back to the server that record which comments have actually been seen (although there is no way to judge if the comment was actually read). It would be a bunch of work though, because the raw log would be enormous and so you'd have to create a special batch processing job to get the view counts split out for each separate comment.
You could probably reasonably approximate this without too much effort.
The number of votes an article has and the number of comments are both going to be strongly correlated with the number of views per comment. So would the mean and range of the comment scores. You could also take the time since the comment was posted into account.
You could have the visible "raw score" for each comment and an invisible "quality adjusted score" that is used to give each user a quality rating.
Comments
The ideal measure would be number of upvotes of a comment divided by the number of views of the comment. That way you wouldn't get punished for making comments late in a thread or on a less well read post. Unfortunately measuring this statistic would require a lot of work to implement.
A lot of work, and it might even be impossible. If someone loads a page with 200 comments, how many of the comments do you judge that he has read?
That said, upvotes per page-load-including-the-comment is probably still better than the raw number of upvotes.
You could hook some javascript to the page's onscroll event. The javascript would send ajax calls back to the server that record which comments have actually been seen (although there is no way to judge if the comment was actually read). It would be a bunch of work though, because the raw log would be enormous and so you'd have to create a special batch processing job to get the view counts split out for each separate comment.
You could probably reasonably approximate this without too much effort.
The number of votes an article has and the number of comments are both going to be strongly correlated with the number of views per comment. So would the mean and range of the comment scores. You could also take the time since the comment was posted into account.
You could have the visible "raw score" for each comment and an invisible "quality adjusted score" that is used to give each user a quality rating.