Hacker News Analysis • Hacker News Board

Re: Hacker News Analysis

#51

Post by Gormo » Mon, May 02, 2011, 4:00 PM UTC

Super_Jambo wrote: post #31

Joakal wrote: post #22

Earlier quoted context omitted.

Your last suggestion seems to make things less democratic understandably, but how would it be implemented?

I've actually implemented a system like this on isdaily (a news site I'm building.) I came to the conclusion that using Karma as a score is a rather bad idea as it assumes that people who get voted up have better judgement: "If you post well you must vote well." But I'm sceptical that this is the case, it's easy to see someone like patio11 has both excellent judgement and a high karma score. But vote power = karma pr…

So in your implementation, you gain voting weight by voting on the posts that have themselves accumulated the highest vote counts?

In order for this to work, you'd have to hide vote totals on all posts and comments, which, as jamesbritt pointed out above, eliminates a feedback mechanism that users may find desirable.

Re: Hacker News Analysis

#52

Post by ugh » Mon, May 02, 2011, 4:09 PM UTC

eru wrote: post #46

ugh wrote: post #18

Earlier quoted context omitted.

There is no way to get a handle on comment and submission quality without content analysis. You need a list of properties that define good comments and good submissions, you need some way to quantify those properties and then you need to invest obscene amounts of work coding those properties. All of this is hard work and full of uncertainty. Different people want HN to be different things, there can’t be one list of…

> This is a Ph.D. thesis worth of work. I wouldn’t try to do it as a hobby. Why not? They won't hurt themselves, and it will be fun. Maybe they will actually learn something.

I wouldn't. I don't object to others trying their luck ;-)

Re: Hacker News Analysis

#53

Post by Gormo » Mon, May 02, 2011, 4:15 PM UTC

bermanoid wrote: post #21

tptacek wrote: post #7

You're working from a different definition of "good comment" than Graham is. The concern with "bad" comments isn't simply that they exist, but that they have deceptively positive scores.

...which is a problem with a crystal clear cause: if you allow purely democratic voting, then you'll end up with really popular crap reaching the top of the pile fairly often. Imagine how useless Pagerank would be as an algorithm if it considered links from every drooler with a domain name to count for the same amount as those coming from the frontpage of Yahoo - that's essentially what you're doing when you tally up…

But a more fundamental question is what we mean by 'post quality'. Right now, the vote total is the only objective metric of post quality; if we're trying to rethink the voting system itself, we need a lower-level set of objective quality criteria if we want evaluate how well a proposed ranking system works.

What are 'crap', 'douchebags', 'fools', and 'sheeple'? Different people have different thresholds for assigning these labels. Until a discussion board establishes quasi-official definitions of these terms, whether by consensus or by fiat, you can't use them to measure anything.

Re: Hacker News Analysis

#54

Post by chalst » Mon, May 02, 2011, 4:19 PM UTC

danvet wrote: post #25

Joakal wrote: post #22

Earlier quoted context omitted.

Your last suggestion seems to make things less democratic understandably, but how would it be implemented?

On a hunch: Run Pagerank (the original thing, i.e. just the stable distribution of the random walk) on the graph of who-upvotes-whom. Then tally upvotes weighted with the upvoters pagerank. It's probably safe to assume that the pagerank of a given user changes rather slowly, so this shouldn't be a too big problem to actually compute (and needs to be update at most every few days). Also, to favour recent history a bit…

Advogato has used an attack-resistant trust metric with similar features; Raph Levien called such systems attack-resistent trust metrics. I'm sure they would do the task you describe successfully. Check out:

1. http://www.levien.com/free/tmetric-HOWTO.html

2. http://www.advogato.org/person/chalst/diary/189.html

Re: Hacker News Analysis

#55

Post by topomorph » Mon, May 02, 2011, 4:39 PM UTC

JCB_K wrote: post #43

I'm surprised there's little under 4000 users. Always thought it was much more.

Just to be clear, that's the number of users with at least one submission in a given month. (Not the number of users with at least one submission or comment, as I sleepily wrote originally, and definitely not the number of distinct visitors.)

Re: Hacker News Analysis

#56

Post by ignifero » Mon, May 02, 2011, 5:17 PM UTC

jbooth wrote: post #50

ignifero wrote: post #32

Earlier quoted context omitted.

That would be a violation of the democratic principle behind social news and would incentivize people to game their scores. Members have different interests and expertise, so having a high avg score doesn't mean they are experts in everything (Abuse of power - OMG this is getting more intricate than i thought)

You could take the log of any score > 1 to reduce the incentive to game the system.. that way someone who relentlessly karma whores won't have that much more influence than a mildly uprated commenter, but both will have more influence than a noob.

but then a mob of noobs could outrank two wise elders.

Re: Hacker News Analysis

#57

Post by masterzora » Mon, May 02, 2011, 8:34 PM UTC

palish wrote: post #37

Perfect example of how graphs can be subtly deceptive: 1) "Variance in number of comments" is visualized as a circle's radius. 2) ... But a circle's area increases as (pi)r^2. That's an exponential increase. A fair visualization would show a linear increase, not an exponential one. EDIT: To be more clear, the goal is "more pixels" = "more variance". Growing the visualization by a radius is inherently unfair. It's pre…

> 2) ... But a circle's area increases as (pi)r^2. That's an exponential increase.

No, no, no, no, no. "exponential" != "has an exponent". Exponential means the variable is the exponent, as in 2^n. The word you're looking for is "quadratic".

Re: Hacker News Analysis

#58

Post by palish » Mon, May 02, 2011, 9:09 PM UTC

masterzora wrote: post #57

palish wrote: post #37

Perfect example of how graphs can be subtly deceptive: 1) "Variance in number of comments" is visualized as a circle's radius. 2) ... But a circle's area increases as (pi)r^2. That's an exponential increase. A fair visualization would show a linear increase, not an exponential one. EDIT: To be more clear, the goal is "more pixels" = "more variance". Growing the visualization by a radius is inherently unfair. It's pre…

> 2) ... But a circle's area increases as (pi)r^2. That's an exponential increase. No, no, no, no, no. "exponential" != "has an exponent". Exponential means the variable is the exponent, as in 2^n. The word you're looking for is "quadratic".

It doesn't matter whether it's unfair by a factor of n^2 or by 2^n. If it's unfair, then it cannot be looked at whatsoever without drawing incorrect conclusions.

Or are you trying to educate me out of good will? If so, thanks.

Re: Hacker News Analysis

#59

Post by masterzora » Tue, May 03, 2011, 8:26 AM UTC

palish wrote: post #58

masterzora wrote: post #57

Earlier quoted context omitted.

> 2) ... But a circle's area increases as (pi)r^2. That's an exponential increase. No, no, no, no, no. "exponential" != "has an exponent". Exponential means the variable is the exponent, as in 2^n. The word you're looking for is "quadratic".

It doesn't matter whether it's unfair by a factor of n^2 or by 2^n. If it's unfair, then it cannot be looked at whatsoever without drawing incorrect conclusions . Or are you trying to educate me out of good will? If so, thanks.

Kind of the latter, really. While the true meaning was clear from context here, it's very easy to make the same mistake in other contexts without the true meaning being so clear. And, unlike a number of other common mistakes, this is one that can really bite you in the ass in some contexts, so it's good to set the record straight before your gluteals are masticated.

Re: Hacker News Analysis

#60

Post by bermanoid » Thu, May 05, 2011, 6:20 PM UTC

danvet wrote: post #25

Joakal wrote: post #22

Earlier quoted context omitted.

Your last suggestion seems to make things less democratic understandably, but how would it be implemented?

On a hunch: Run Pagerank (the original thing, i.e. just the stable distribution of the random walk) on the graph of who-upvotes-whom. Then tally upvotes weighted with the upvoters pagerank. It's probably safe to assume that the pagerank of a given user changes rather slowly, so this shouldn't be a too big problem to actually compute (and needs to be update at most every few days). Also, to favour recent history a bit…

Whoops, missed a lot of responses here, sorry about that!

Yes, vanilla Pagerank (where the "sites" are accounts and "links" are upvotes) is the first thing that came to my mind, mainly because it's been pretty well battle tested; running it on the comment graph is actually a much simpler problem than running it behind a search engine because you don't need the additional refinements based on search keyword. A comment's Pagerank is, by itself, enough to set an ordering.

If that's too much work, though, even a simple vote weighting based on some function of the upvoter's karma would be a decent approximation. Say every user's votes had an impact based on some sigmoid function of their karma (maybe a Gompertz function?), tuned so that the plateau is hit for the top 10% of users or something like that.