12

I regularly scan the Stack Exchange front page for interesting questions. It looks like among all the sites Code Golf has been overrepresented for a long time, and right now it got kind of weird: five out of the top ten (and four out of the top five!) questions on the front page are from Code Golf. Is there some weighting that's gone wrong, or is Code Golf really by far the most popular of the Stack Exchange sites?

1
6

I think Code Golf just lends itself very well to having hot questions. Answers are taken into account when calculating the hot network questions score:

(MIN(AnswerCount, 10) * QScore) / 5 + AnswerScore
-------------------------------------------------
         (QAgeInHours + 1) ^ 1.4

On Code Golf, there will generally be many answers because each language gets its own answer. So, every answer up to a maximum of 10 answers will make sure the posts from Code Golf get a little boost in hotness score. Add in some active voting and you have a recipe for hot posts. So the weighting isn't wrong, these posts really are the hottest at the moment. There are other sites that often reach HNQ because of the amount of answers and votes a post gets, like Worldbuilding or The Workplace.

To prevent overrepresentation though, each site can only have a maximum of 5 posts in the HNQ list at a time. And sites can ask Stack Exchange to lower that number, for example when they feel the amount of posts in the HNQ at any time put an unwelcome burden on community moderation work on their site. But this is something that needs to be discussed by the people active on that site, not by people that are just getting tired of seeing the site's questions on HNQ.

1
  • 1
    Oh right, so the "arbitrarily rewarded" tool tip text on the links on the front page is maybe a bit misleading if the function is public knowledge. – l0b0 Apr 19 at 19:53
7
+50

What's going on here boils down to two things:

  • The questions on that site invite many answers per question
  • The community is very active with voting

This leads to questions with high scores in the formula that determines which questions appear in the Hot Network Questions list. If we wouldn't have a maximum of five questions per site in the list, there surely would be more of them in the top regions. Some sites are penalized, which is an old way of making sure not too many questions would appear in the list, but I guess that wouldn't matter here. See, not many people are browsing the Hot Network Questions the way you do; most traffic comes from the widget in the right sidebar, and the questions shown there are randomly chosen from the top 100; the top 10 isn't shown more often than the bottom 10.

Actually, I'm not even sure Code Golf is the most 'popular' site when it comes to Hot Network Questions. A few years ago, it used to be Worldbuilding; here is some research I did back then. Code Golf was already number one when it comes to % of questions which become hot. When I have time, I'll have a look at updating that list with figures from 2020.

7
  • 2
    Like the OP, I frequently check the HNQ list. I can confirm that Code Golf and The Workplace are definitely amongst the highest represented most of the time, with Code Golf probably being the top one. – 41686d6564 Apr 19 at 8:58
  • as time goes by I am getting more and more inclined to believe that the right way to address vast majority of HNQ issues is, paradoxically, to turn off penalizing at (the only site where it makes a solid impact) Stack Overflow. This way we'd have all the worthy issues (currently experienced only by smaller sites) exposed at SO and - which is most important - have them raised at MSO, which is the only per-site meta capable of making the company listen and improve things. This way would allow to naturally reach proper tuning of HNQ system parameters via target community feedback – gnat Apr 19 at 15:00
  • ...when Atwood penalised SO very long time ago, he had pretty good reasons for that. However, even back then his approach was maybe not most optimal, kind of throwing the baby out with the bathwater by suppressing useful feedback from SO audience. But now that any site has hard stop at 5 questions, this tweak no longer makes sense and completely turned into a cure worse than decease. Silencing feedback certainly makes things feel more comfortable for management but what we get "in exchange" is a poorly balanced system (since nobody knows what is really worth improving and what isn't) – gnat Apr 19 at 15:22
  • 1
    fwiw, another way the HNQ list generates traffic is when you use its RSS feed. I do this. It is pretty dependable as a boredom relief. – Ross Presser Apr 19 at 17:06
  • 1
    "most traffic comes from the widget in the right sidebar, and the questions shown there are randomly chosen from the top 100" correct but Code Golf and The Workplace always have 5 questions in the top 100, and it's much easier for those sites to get their questions into that list than it is for smaller sites or sites where questions get only 1 answer. Upvoted Code Golf questions almost "automatically" will enter HNQ if they're not too hard to answer and the max of 5 hasn't been reached yet. – user1271772 Apr 21 at 21:47
  • @user1271772 "Code Golf and The Workplace always have 5 questions" Code Golf, maybe (I haven't checked) but The Workplace doesn't always have 5 questions. It's still one of the highest represented though. – 41686d6564 Apr 22 at 5:57
  • @user1271772 that was mainly aimed at the assertion in the question 'Code Golf has 5 questions in the top 10' - my reply was that it doesn't matter where they are in the list. – Glorfindel Mod Apr 22 at 7:23
3

"is Code Golf really by far the most popular of the Stack Exchange sites?"

It's not the most popular.

Consider these numbers which you can see here:

Sites with most questions/day in the last 14 days:

  1. StackOverflow: 6712/day

  2. Mathematics: 567/day

Code Golf has 3.6/day

Sites with most visits/day:

  1. StackOverflow: about 11 million/day

  2. SuperUser: about 761 thousand/day

Code Golf has about 6000/day

Sites with most users:

  1. StackOverflow: about 14 million

  2. SuperUser: about 1 million

Code Golf has about 81 thousand.

"Are Code Golf questions overrepresented on the front page?"

Rene wrote this app which regularly shows Code Golf along with some other sites like World Building and The Workplace that frequently have the maximum of 5 Hot Network Questions (HNQs).

The other answers have given a formula that approximates how the HNQs are determined. Keep in mind that the real formula for this is not publicly available[1], but the main point is true, which is that the number of answers on a question has a big impact.

With Code Golf (and some of the other sites that get a lot of HNQs), the questions do not have just "one authoritative answer" but they invite several people to answer. At Code Golf, right now there's 124 answers across their 5 Hot Network Questions! This makes it much easier for them to get in the HNQ list than sites where, for example, Mathematica.SE where people tend to ask how to do something in Mathematica and there's sometimes only one answer to the question.

Is Code Golf over-represented in the HNQs list?

Along with others like World Building and The Workplace, their questions do tend to get into the HNQ list easier. Perhaps the part of the HNQ formula which contains a multiplication factor by the number of answers to the question, can be damped by the site's overall question-to-answer ratio! I don't see that causing any serious problems that are bigger than any of the problems that currently exist. But keep in mind that there will never be a "perfect" formula for scoring the "hotness" of questions. Another foible with the formula is that the chances of a question appearing in the HNQ list decays exponentially with the time that's passed since the question was asked, which makes it unfair for sites where very difficult questions get asked (where the average time to get an answer is longer than 3 days, for example).


[1] My answer shows some examples of the HNQ formula failing: Questions with a lower score from the formula, appear as HNQ over questions with a higher score, despite taking into consideration all caveats of the formula such as the maximum of 5 questions/site and the known penalties (site's that appeared to be penalized in one experiment seemed not to be penalized in the next experiment, for example), etc.

You must log in to answer this question.

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