27
$\begingroup$

I know that the Hot Network Questions are chosen by some algorithm, but I was wondering if there is any configurability in that?

If not, I would like that as feature request:

It would be very good if site-moderators (for graduated sites) could specify certain limitations, i.e. either an inclusion or and exclusion filter cirteria which is applied on top of what the HNQ algorithm does.

f.e., one could define that

  • only questions posted by a poster of rep > limit could be chosen
  • a specific tag has-to/must-not be present
  • the post must be at least XY days old (giving time to VTC is first)
  • etc.

Such a mechanism (customized for Puzzling) would certainly help to avoid the repeating problem of the HNQ inviting exactly the wrong type of content to the site. It is very hard to VTC & DV "bad" content, if exactly those questions get a flood of "me-too" facebook-type "likes" within the first couple of hours.

I could imagine that other sites could benefit of such customization as well.

$\endgroup$
  • 3
    $\begingroup$ HNQ drives a lot of traffic here, is there a reason for limiting it? $\endgroup$ – user2052 Jun 19 '16 at 3:06
  • 10
    $\begingroup$ @Snowman Not limiting in the sense of reduction but in the sense of steering what content gets to HNQ and therefore attracts what kind of traffic. $\endgroup$ – BmyGuest Jun 19 '16 at 8:37
  • 5
    $\begingroup$ @Snowman Bigger isn't necessarily better. If the traffic brought by HNQ is people who are going to post inane riddles with nonsensical "guess-what-I'm-thinking" answers or other poor-quality content, we don't want that traffic anyway. $\endgroup$ – GentlePurpleRain Jun 22 '16 at 15:10
  • 2
    $\begingroup$ must-not-be-present = riddle $\endgroup$ – Dan Russell Jul 26 '16 at 19:34
12
$\begingroup$

This answer explains how the HNQ algorithm works. The basic math is as follows:

$$ \frac{(min(\text{AnswerCount}, 10) \times \text{QuestionScore}) \div 5 + sum(\text{AnswerScores})}{max(\text{QuestionAgeInHours} + 1, 6) ^ {1.4}} $$

This means the two main things that drive the HNQ score up are QuestionScore and AnswerCount.

On Puzzling, a multitude of answers generally indicates a poor puzzle. Either it is not well-defined, and thus open to interpretation, or it is too ambiguous, and has many potential answers that could fit.

Here are some numbers to show how this works (these all assume a question age of < 6 hours):

$$\begin{array}{c|c|c|c} \text{QScore} & \text{ACount} & \text{AScores} & \text{HNQ score}\\\hline 10 & 3 & 12 & 1.465\\ 10 & 4 & 16 & 1.953\\ 20 & 5 & 20 & 3.256\\ 40 & 5 & 20 & 4.884\\ \end{array} $$

In general, the Question Score is much higher than the Answer Count, so I would venture to say that it is the single-most-influential data point in the calculation above.

So to prevent "the wrong type" of question from getting into HNQ, downvote1 them aggressively!
If you see questions that you think are of inferior quality, use your downvotes. That's what they're there for.

Some people may not be aware that downvoting a question costs you nothing. You lose 1 reputation for downvoting an answer, but question downvotes are free, so if you don't like what's appearing on HNQ, downvote1 it!

Oh, and did I mention that you can downvote1 inferior questions?

Downvote - Downvote - Downvote - Downvote - Downvote - Downvote


1 NOTE: Especially with new/low-rep users, it can be very helpful to post a comment when you downvote, explaining why you think the question is not suitable for this site. (Thanks, @BmyGuest)

$\endgroup$
  • $\begingroup$ One loses 2 rep points by having its answer downvoted, but actually loses only 1 point by downvoting an answer. $\endgroup$ – Anton Jul 28 '16 at 12:37
  • $\begingroup$ @Anton, Of course; you're right. I've updated my answer. $\endgroup$ – GentlePurpleRain Jul 28 '16 at 14:26
  • 1
    $\begingroup$ But it is exactly because a question score is higher that downvotes don't impact the HNQ score by nearly as much as an added answer. Consider a question with 10 score and 1 answer. If only 3 more users add answers to it, it's now already equivalent to a question with 40 score, that is, 30 more upvotes! $\endgroup$ – ffao Jul 29 '16 at 4:08
  • 3
    $\begingroup$ Also, I don't think you have covered the option of downvoting questions of inferior quality in this answer. $\endgroup$ – ffao Jul 29 '16 at 4:15
5
$\begingroup$

With recent discussions and active attempts to handle the declining riddle quality on the site, I thought I'd also resurrect this question too, as I think it's related (since broad low quality riddles => lots of guesses => HNQ)...


As GentlePurpleRain points out, the algorithm to select posts for HNQ is as follows:

$$ \frac{(min(\text{AnswerCount}, 10) \times \text{QuestionScore}) \div 5 + sum(\text{AnswerScores})}{max(\text{QuestionAgeInHours} + 1, 6) ^ {1.4}} $$

Issue: As noted, this tends to promote exactly the wrong puzzles from PSE, as it's usually broad, low quality questions that gather many responses in a very short space of time.

Assumption: The SE devs won't be willing/able to change this algorithm on a per-site basis.

Challenge: How could the algorithm be modified across the entire network such that it doesn't make it worse for other sites, whilst still helping to somewhat mitigate the issues we see here on PSE?

Suggestion: Ask the powers-that-be to modify the algorithm slightly, such that it has little to no impact on other existing sites*, but gives us a little more control over keeping low quality puzzling content out of the HNQ.

Specifically, I suggest that we modify $\text{QuestionScore}$, in the algorithm above, such that it is calculated, not as a simple score, but as:

$\text{QuestionScore}$ = $\text{QuestionUpVotes} - 3 \times \text{QuestionDownVotes} - 5 \times \text{QuestionVTC} $

Reasoning: All this change does is to allow downvotes to have a higher impact in preventing a question from qualifying for the HNQ list, and to allow our experienced, 3000+ rep users to have their close votes help even further. As it currently stands, there's a feedback loop where a broad/low quality puzzle hits the HNQ quickly, gaining it more drive-by upvotes than can be balanced out by downvotes from the community, which in turn keeps it "hot". This proposal, I think, would help to slow/reverse that feedback loop on poor quality stuff, whilst neither blocking "good" content nor impacting what qualifies as "hot" on other sites in the network.

* In fact, it would arguably have a slightly positive effect as it would help filter out more controversial posts in favour of universally praised ones.

$\endgroup$
  • 1
    $\begingroup$ I think the HNQ problem is driving a lot of the mediocre content on the site and I wholeheartedly welcome any attempts to try to address it somehow. Thanks for adding your 2 cents here. Maybe one of these ideas will land and make a difference. $\endgroup$ – Dan Russell Aug 29 '16 at 2:38
2
$\begingroup$

This answer gives another aspect of the HNQ mechanism, namely that questions with MathJax in their titles are excluded from the HNQ list.  That's in response to a question about $\;$ being appended to a title.  \; is MathJax for a somewhat-wider-than-normal space, and so is invisible if nothing follows it.  ~ is MathJax for a normal space, so, presumably, $~$ would work as well.

Any 2K user (or a <2K user with two 2K approvers) can edit an undesirable question to add one of the above strings to the title.  This can be done in a matter of minutes (much quicker than VTCing).  Of course, it can be reversed just as easily.  But maybe the sort of person who writes (or wants to encourage) such questions won't understand the trick and won't be motivated to mitigate it.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .