The problem with this that I foresee is that it seems rather impossible to predict actual views of a comment. I heavily doubt that most viewers of the comment page will read most of the comments. So this will create a positive feedback loop that just creates a large divide between comments that (somehow, probably mostly arbitrarily) start off more popular than others; comments with high "rates" will be pushed to the top, and will be seen most of the time someone views the page, but comments with low "rates" will be pushed to the bottom, meaning they won't be seen most time a page is viewed, and so the estimate of their rate will be biased downwards.