Timeline for In “network hot” questions formula, discard answers when voting evidence indicates that these are not good data points
Current License: CC BY-SA 3.0
27 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 20, 2017 at 10:31 | history | edited | CommunityBot |
replaced http://meta.stackexchange.com/ with https://meta.stackexchange.com/
|
|
| Apr 24, 2014 at 13:52 | history | edited | CommunityBot |
Migration of MSO links to MSE links
|
|
| Mar 7, 2014 at 6:05 | comment | added | gnat | David, I would appreciate if you do similar analysis for another approach to "approximate" this feature request: Is it technically feasible to have 2-3 aging factors for hot questions at Programmers? At the moment, I am primarily trying to figure if particular technical change would involve blocking performance implications, like it happened with this request | |
| Nov 14, 2013 at 6:21 | comment | added | gnat | bounty awarded for providing detailed clarifications on the current implementation and for thorough analysis that helps to better shape the requested change | |
| Nov 14, 2013 at 6:21 | history | bounty awarded | gnat | ||
| Oct 23, 2013 at 17:27 | comment | added | gnat |
@GlenH7 I mean, in current analysis, these these impact values of NumAnswers -> QScore * NumAnswers / 5 part of the score - this needs adjustment - though I don't think it will change much, for the reasons explained in prior comments
|
|
| Oct 23, 2013 at 13:41 | comment | added | user194162 | @gnat it's worth noting that the 10th -> final answers are pretty much chronologically ordered since they never received any upvotes. Based upon the algorithm clarifications, I don't know that those answers contribute anything to the collider score. They're shunted from the QScore multiplier and they have zero votes so no contribution to SumAnswers | |
| Oct 23, 2013 at 9:16 | comment | added | gnat |
...age-drag effect should go away (ie should stay at 1 through all the time), but again, it's not that far from it now (between 0.8 and 1.3) and, which is most important, it doesn't really impact the core point made about "relative effect of QScore * NumAnswers / 5 versus SumAnswers". Thing is, this relative effect simply doesn't depend on the age drag: both values are scaled so that drag effect, whatever it is or isn't, simply disappears when one divides them to get relative effect
|
|
| Oct 23, 2013 at 9:16 | comment | added | gnat | ...Regarding analysis answer, it's probably worth noting that, although somewhat outdated, it still feels in the ballpark. For first 3 hours, the example question had up to 12 answer, making calculation pretty close to those that would be there with 10 posts limit. And even after that, number of answers has been up to 16 which is, again, not too far from 10. Denominator, denoting... | |
| Oct 23, 2013 at 9:16 | comment | added | gnat | David, now that the feature request and implementation notes have been updated to reflect your clarifications, does it make sense to also update decline justification? (BTW your note #4, on denormalized field was spot on - well done!) I can open a separate MSO discussion for that if it feels too big to discuss in comments... | |
| Oct 8, 2013 at 17:25 | comment | added | gnat | thanks for understanding! in my experience isn't ideal is very soft charachterization | |
| Oct 8, 2013 at 17:14 | comment | added | David Fullerton StaffMod | @gnat ah, I see what you mean. I agree, the current formula isn't ideal. It's on my list to look at more when I get some time. | |
| Oct 8, 2013 at 14:43 | comment | added | gnat | found the reference for best community effort I mentioned: diamond/10K-only link to deleted post. 9 DVs on the question and up to 6 DVs on the answers. Yeah we try to do our best to keep site professional but we can only do so much | |
| Oct 8, 2013 at 14:42 | comment | added | user194162 | To continue the example, let's say I downvote all 6 of the poor answers. The 6 answers have contributed +60 to the collider value while my DVs contribute only -6 to the collider value. So the poor answers have a disproportionate effect in attracting attention to the question. Even if gnat comes along and DVs the 6 along with my DVs, that's still only -12 to AnswerScore vs +60 from QScore*Answers. Giving that DVs are comparatively rare and there is minimal effect from them, the community can't rein in a poor question that's hit the collider. | |
| Oct 8, 2013 at 14:31 | comment | added | gnat |
David, I specifically refer to "we only look at the score of the top 3 answers" (hope this part is really gone and is not just lost in edit). And anyway, did you do the math (using eg this example): adding an answer (any answer) increases hotness by 83/5=16, how many downvotes would you expect to compensate for that? Note BTW that the best community effort I've ever seen at Programmers was like 5-6 downvotes, and it was barely sufficient to overcome sympathy upvotes piling on, thanks to Trouble with popularity
|
|
| Oct 8, 2013 at 14:29 | comment | added | user194162 | @DavidFullerton - Yes, down voted answers reduce the AnswerScore, but they don't degrade the QScore portion of the numerator. QScore is multiplied by number of answers, which means that a poor answer (up to the tenth one) feeds into the final score. For example, a question with 50 upvotes will pick up 10 collider points for every answer (up to 10) that it receives. The 5 or 6 poor answers on the question will still add 50 - 60 collider points in this example. | |
| Oct 8, 2013 at 14:25 | comment | added | David Fullerton StaffMod | @gnat I'm not sure what you mean. When users downvote answers, AnswerScore decreases, which decreases the overall hotness. See the formula here: meta.stackexchange.com/a/61343/146719 | |
| Oct 8, 2013 at 14:06 | comment | added | gnat |
David I wonder if you realize that the way how you describe the formula ignores community feedback in quite a brutal way? Just think of what happens when users downvote 5th, 6th, 7th low quality answer - formula keeps stuffing the hotness score with QuestionScore/5 no matter what, only making matters worse, increasing question exposure with all the associated risks of getting more low quality answers (just what has been observed in referred studies)...
|
|
| Oct 1, 2013 at 21:13 | comment | added | gnat | regarding views, in the context of this feature request, it is purely tangential whether formula counts these or not; request explicitly factors views out: "several hundreds views from collider do not come close to thousands views that come when question becomes fairly popular from outside..." (and yeah, at Programmers, we even tested and discussed this) | |
| Oct 1, 2013 at 19:55 | comment | added | gnat | wrt "denormalized field" to keep evidence-based score, I never ever imagined having stuff like that for every question - it just does not make sense as there are only 100 questions in hot list anyway (half of them at collider). Just 1) select 200-300 top-hot questions using current formula as approximation 2) recalculate score for these using voting evidence and 3) select top 100 by re-calculated score. Simple as that | |
| Oct 1, 2013 at 19:18 | comment | added | Mad Scientist | There are a few things I'd change about the formula: A hard cutoff after x days, scaling not only for traffic but also for voting behaviour and removing CW questions (due to the few sites still using CW for soft questions). I think all of those have feature requests already. I'd like to play around with the algorithm to make more specific suggestions and add some hard data to support them, but as far as I see the necessary data is just not accessible to regular users. | |
| Oct 1, 2013 at 16:27 | comment | added | user194162 | Looks like this is the current formula: meta.stackexchange.com/a/61343/194162. Thanks for updating your other answer! | |
| Oct 1, 2013 at 13:42 | comment | added | gnat | I think it's also worth mentioning that even cap at 10 answers doesn't really address the issue of using "bad data points" in the formula - "Even with tweaked formula, stuffing 8 useless, zero-score answers into +50 question would have the same effect as giving 80 upvotes to answers. At +200 question, this would be like giving 320 (over three hundreds!) upvotes to answers... and there's even a real, recent example for that..." | |
| Oct 1, 2013 at 13:13 | comment | added | gnat | regarding the "more open-ended feature request" you mention, does this one fit: Don't let questions stick to the top of the hot questions list forever? Per my reading it points to the issue almost precisely as you describe: "Hot questions stay at the top of the supercollider for too long" | |
| Oct 1, 2013 at 13:11 | history | edited | David FullertonStaffMod | CC BY-SA 3.0 |
edited body
|
| Oct 1, 2013 at 13:04 | history | edited | Grace NoteStaffMod | CC BY-SA 3.0 |
Need... colors... badly...
|
| Oct 1, 2013 at 12:57 | history | answered | David FullertonStaffMod | CC BY-SA 3.0 |