7

The following question has a score of -1 votes, and 4 close votes: https://superuser.com/questions/419400/where-can-i-buy-micro-express-microflex-23b-in-europe

I do not think it should be included in the Top Questions list. Also, notice the [closed] question above it! That being said, I think the algorithm needs an update, at least to reflect that some questions are poor fits for the site:

  1. If the question has a score greater than N votes, include it, else,

  2. If the question is closed, do not include it.

  3. If the question has more than 3 close votes, do not include it.

  4. If the question has a score of less than 0 votes and if there are no answers, do not include it.


Screenshots:

Top Questions

419400

  • 1
    If you listed some things you wanted to change with the algorithm, then it would be a feature-request. But as it is, reads like support – random May 2 '12 at 1:05
  • @random Thanks. I could come up with a few suggestions. – iglvzx May 2 '12 at 1:05
5

Example:

(I totally disregard that age isn't just integers. It'll probably only get worse for very recent questions)

Works as designed. It's just the algo that sucks.

  • Very recent questions are hot by design
  • Getting some views is better than some score and answers
  • Question score doesn't matter if there are no answers

Unsurprising, since the algorithm was designed with SO's traffic and voting and answering behavior in mind.


Let's see...

screenshot of two very different questions

  • ln(13)*4+(1*1/5)+1 divided by 1.35 is about 8.5
  • ln(79000)*4+(326*2/5)+679) divided by (33-((32-20)/2)^1.5) is about 31.7

Now that one I don't get... unless Qage and Qupdated are measured in minutes, resulting in hotness scores of 0.55 and 0.54, respectively, making younger question much hotter using this scoring. Clearly, the rules aren't made with our rather slow uptake of question scores and views in mind.


Update 2

In case you wondered about the following:

Screenshot of boring 0 score question above Carmack question

These are the numbers (Qage and Qupdated really are in minutes):

  • Three hours old, 0 score, 0 answers, 16 views:

    (ln(16)+(0*0/5)+0)/(180-((180-180)/2)^1.5) = 0.15403

  • Five days old, 93k views, 437 score, three answers with 861 combines score:

    (ln(93729)+(3*437/5)+861)/(7680-((7680-7410)/2)^1.5) = 0.150384

Clearly, age is way overemphasized in the algorithm. That it doesn't have minimum values for score, answer count, or views is also a problem.

  • From my experience, "bad" questions attract just as many views as good ones (and definitely more than the average ones). As a start, views should only be taken into account if the post has no negative score. Especially when there are no answers. – slhck May 2 '12 at 12:01
  • As a contrived example, a 55 views topic with no answers and arbitrary (e.g. negative) score is slightly hotter than a 12 views, 5 score topic with 2 answers, scored +3 and +1 created at the same time. – Daniel Beck May 2 '12 at 12:02
  • Unit of time is mostly relevant for edits, when Qupdated != Qage, but because of the +1 zero-division protection, it'll also affect recent topics very strongly (1+1 : 1 is very different from 60+1 : 60). – Daniel Beck May 2 '12 at 19:14
2

There is a specific formula which determines the hotness score.

(log(Qviews)*4) + ((Qanswers * Qscore)/5) + sum(Ascores)
--------------------------------------------------------
((Qage+1) - ((Qage - Qupdated)/2)) ^ 1.5

I'm really not sure how that one got selected. Could be a bug.

  • 1
    Easy: If there are no answers, the score doesn't matter. What's left is the views divided by a very small divisor, as with all very recent posts high up on the (not so) hot list. – Daniel Beck May 2 '12 at 8:01
  • @DanielBeck that doesn't really make sense. If Qanswers is 0, then then that whole term of the equation is 0, and sum(Ascores) is also 0, so the overall score should be lower than if there were answers, because the numerator is going to be smaller. – nhinkle May 2 '12 at 8:24
  • Check out the screenshot with three very recent questions being considered hot. Due to their very little age, it doesn't matter that they get all their hotness from their views. I just explained why the negative score isn't relevant, since the user didn't highlight the other 0 score or 0 answer topics in the screenshot. – Daniel Beck May 2 '12 at 9:04
  • 1
    @DanielBeck ah, I see better what you mean now. Thanks for showing that. The hotness algorithm does tend to be... pretty bad. – nhinkle May 2 '12 at 16:32

You must log in to answer this question.

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