The problem with using karma is that it is affected by the regression to the mean. As more "average" people come in, more "average" people will give them karma. Solve that problem, and you might have a solution to the whole problem (I, for one, am not smart enough for that ;).

Alternatively, perhaps we can let users select (follow) a set of like-minded people and make the up-votes of the people they pick weigh more on a "customized" front page.

But then people may have not like having their up-votes be so transparent..

Maybe some mix of karma, account age, and the number of upvotes received from first-year members to give a person's votes weight.

If you really wanted to make the site more like it was, you could look at the voting history of people here at some distant time in the past and find other members whose votes highly correlate with the old timers. Some people just quietly upvote, so they don't have much karma. If you want the upvoting to be more like it used to be, award people who vote similarly like how older users used to vote.

It would help take care of the effect where older uses leave due to changing interests etc.

Note: I don't actually think it's a good idea. I think just figuring out _what_ you want exactly, in vaguely quantifiable role, is sort of a hard problem. I suspect a lot of people have different 'golden ages' in mind when they think of HN.

When I came here originally it was pretty startup heavy, which was sort of remote from what I was up to at the time. I liked it when it got a little away from startup business related stuff (my perception) and toward more generic technical issues. My point is even if everyone agreed that ait has gone done hill (an open question), not everyone might agree on what it should be like. It's not a democracy of course, but even as an individual for my own preferences that seems as hard as the technology issues.

What about using baysian filtering? We already have a lot of data on good and bad comment styles, baysian filtering could give us an indication if a comment is violating HN guidelines.

I am not so optimistic about submissions because the data we currently have is tainted by submissions we don't want. However by using the technique you have described, we could probably achieve better results.

This problem applies to all communities using votes to sort user-generated contents: HN, Reddit, StackOverflow, Quora, etc.

I think something like PageRank can really help. The same need is to rank things based on a directed acyclic graph. The only difference is we have multiple kinds of vertexes: users, submissions, comments.

Hm... Story upvotes modulated by users Karma might be good enough. It's an interesting concept...

Stackoverflow operates under this principle, more or less, but it doesn't seem to entirely stop the slow decline.

But the interesting issue here is that those who write interesting commentary may not be good at discerning good quality. (according to whatever standard we're holding)