34

The Hot Network Questions page displays a constantly changing list of 100 questions, since it deals with time since recent activity.

But are changes in this list ever recorded anywhere for posterity (or data mining, or to use in studies like 1, 2, 3, 4, 5, 6 etc)? Can one, perhaps someone with sufficient data access privileges, look at a certain question and see when it was hot, in the form of a graph or something?

As far as I can tell, history of this ever-changing list isn't stored anywhere:

We currently do not record anything about when a question enters/leaves the hot question list. The list is calculated on the fly each time (and heavily cached)...

This is a feature request to add an audit log or something somewhere for that.

I don't know how often is the list actually updated, but for data study purposes I think it would be okay to record changes once in an hour or two or, if even this would be too difficult, at least once a day would probably be acceptable.


Note there is a similarly looking question:

It only looks similar (partly because I simply modeled wording above after it:) but is actually different.

As opposed to current request which asks essentially for timeline of changes in a list of 100 elements (question IDs), older one is about much much more detailed data, namely about recording hotness score of all (!) questions all (!) time, if I read it correctly.

Not only it would be much harder to maintain and analyze such data, but its very usefulness became quite questionable after shuffling was introduced at sidebar, meaning that hotness score became simply ignored to a large extent.


Another older question which is indeed strongly related to current one:

The difference is that according to accepted answer from SE developer, that older request somehow involves fairly cumbersome (real time?) tracking of vote counts, which is totally irrelevant here, as current request is limited strictly to "timeline of changes in a list of 100 elements", possibly recorded "once in an hour or two or... at least once a day".

4
  • 2
    We currently do not record anything about when a question enters/leaves the hot question list. The list is calculated on the fly each time (and heavily cached), so this will be a feature-request to add an audit log or something somewhere for that.
    – Kasra Rahjerdi StaffMod
    Commented Aug 22, 2014 at 18:19
  • 1
    @KasraRahjerdi thanks a lot! I retag and retitle per your clarification
    – gnat
    Commented Aug 22, 2014 at 18:20
  • related (not a duplicate): Does Stack Exchange Track Clicks to “Hot Network Questions”
    – gnat
    Commented Sep 18, 2014 at 16:44
  • changes listed here address concerns I raised here sufficiently to consider it a duplicate
    – gnat
    Commented Apr 26, 2019 at 9:46

3 Answers 3

4

As of February 28, 2019, questions that enter the Hot Network Questions list will get an entry in the revisions/post history table, with type 52. It's visible on the site itself in the revisions list:

and you can use the API (and probably SEDE as well) to search for these events.

There is (as of yet) still no official record of questions leaving the Hot Network Questions list.

7

I wasn't going to go out on a limb but since a comment has been made...

It would amount to little more than a tool to feed speculation and paranoia...

...I would like to address that.

If it would be so, speculation and paranoia would probably be already here, because said "tool" is available for anyone for about half year now, as explained in related MSE post:

Script that automatically will send you the amount of hot questions that you want as often as you like...

Given above, reluctance to establish proposed audit log (even for internal use) seems to only deprive Stack Exchange team of an opportunity to learn more about content that goes to site main page and sidebar. Anyone else can obtain and use this information.

Overall, attitude in quoted comment reminds security through obscurity, not quite a productive approach.

https://brooksdebates.files.wordpress.com/2008/10/head_in_the_sand.jpg

2
  • 3
    Well, haven't they made it clear enough to you yet that they don't friggin care how the hotlist works in practice? ;-) (Still, don't let that stop your efforts, you are a shining light at the front of fighting the apparently hopeless struggle to repair this completely flawed hotlist "feature".) Commented Mar 3, 2015 at 18:20
  • 1
    @ChristianRau for the sake of proper attribution, I mostly build upon ideas originally laid out by another user
    – gnat
    Commented Mar 3, 2015 at 18:30
3

This still isn't going to give you what you want. I say that even as I realize I don't actually know what you want, because you didn't state it anywhere. Be specific!

But right now, it doesn't really matter; as you note,

its very usefulness became quite questionable after shuffling was introduced at sidebar, meaning that hotness score became simply ignored to a large extent.

This same change is going to hurt the usefulness of knowing what's in that list of 100 questions: most views come from the little sidebar that shows up when viewing a question. Now, you might assume that the random selection is fair and each question in the master list gets an equal share of views via the sidebar - but that's getting into dangerous territory, and if nothing else becomes completely useless if we ever start to customize this list according to user preferences (a much-requested feature).

I suspect what you really want to know is:

  1. How many people are shown a given question via the hot lists?
  2. How many people click through?
  3. What else do they do once they've clicked?

We currently track #2 and #3, with the caveat that we don't track the specific question clicked on (it's possible to obtain this information by cross-referencing multiple logs, but that's a huge inconvenience). We do not track #1.

Tracking #1 would result in a huge amount of data; I'm not sure if it is feasible. We do sample some other high-volume events like this, so that might be an option.

Personally, I would be more interested in fleshing out #2 to include the question clicked: we did track that at one point, but appear to have dropped it. It's possible to make an educated guess with the help of the proxy logs, but that's not really suitable for anything more than occasional one-off analysis of odd behavior.

As it stands, we're able to answer questions that rely solely on the data exposed by #2 and #3, which is to say: broad behavioral patterns that might vary by site or user class. But we cannot easily answer questions as to whether specific types of questions on a given site influence behavior of viewers coming in via the hotlist (although if we ignore the hotlist, it is possible to analyze user behavior according to both user and question class). As long as there is a specific question.


Update: I asked the devs responsible for our event tracking system to look into this a while back - the amount of data involved here would not be feasible to collect and process in our current system. These are shown to everyone, and the number of views is extremely high.

I still do not believe your original request - tracking the presence of questions in the list regardless of who they're shown to - is useful. However, if you're satisfied with a sampling, you can satisfy this request yourself - just scrape the network list a few times every hour and compile the results. There's no reason to permanently log this on questions.

14
  • 1
    regarding specifics, request lists (six) examples of studies where this data is expected to be useful. Off top of my head I can tell that I expect it to help in learning about questions that "stick" in the list for possibly too long and about how different sites are represented in the list. When studying specific questions, I also expect requested data to complement details currently provided in question timeline and revisions...
    – gnat
    Commented Aug 24, 2014 at 16:52
  • 2
    ...I also expect it to help us learn how much of an issue is getting inappropriate questions in the list and how often it happens that close votes of site regulars get wiped out by "popularity views". Etc. On a more general note, I am sort of appalled that stuff considered important enough to represent site main page doesn't even have a rudimentary log - and that request for such logging needs some additional justification. Correct me if I'm wrong but this sounds audit 101 to me
    – gnat
    Commented Aug 24, 2014 at 16:53
  • 1
    ...probably worth noting that request doesn't ask about anything massive. Just a table with two fields, one for question ID, another for timestamp (or three fields if you prefer to store changes - third field being boolean to denote whether question enters or leaves the list). 100 questions in the list - 36,500 rows in a year if logging is done once a day. Data like this can be stored and processed with Excel on an average home PC
    – gnat
    Commented Aug 24, 2014 at 17:03
  • Again, what you're specifically asking for here won't help without also logging how many actual views each item gets - and who is viewing them. That I support. Knowing that something blinks in and out of the hotlist for a week without knowing if anyone saw it... That I'm not convinced is useful.
    – Shog9
    Commented Aug 24, 2014 at 17:08
  • 2
    I would like to make it abundantly clear that yes, the way how I expect to use that log has nothing to do with views. How long question sticks, how are different sites represented, how many of the listed questions get 3-4 close votes - all of this can be analyzed just fine without any data about views. In a perfect world, I'd love to have also views (oh-kay, feel free to add that third-fourth field to that table to record views) but frankly I think I could easily live without
    – gnat
    Commented Aug 24, 2014 at 17:15
  • Without views, there is zero point to implementing this. It would amount to little more than a tool to feed speculation and paranoia, more of the same "hot list boogieman" fear of the outside nonsense we already see. I'm interested in information that helps us make things better, not ghost hunting.
    – Shog9
    Commented Aug 24, 2014 at 18:16
  • do I understand it correctly that 1) you adamantly against implementing it without views and that 2) with views this falls into "Tracking #1 would result in a huge amount of data" in your answer? (ie that's simply a justification for status-declined)
    – gnat
    Commented Aug 24, 2014 at 18:21
  • Whether or not #1 can be implemented boils down to how clever our devs get when faced with vague requirements...
    – Shog9
    Commented Aug 24, 2014 at 18:46
  • understood, thank you. Good luck in gathering data about meta effect impact on Stack Overflow questions
    – gnat
    Commented Aug 24, 2014 at 18:48
  • 2
    I'm very disappointed by this answer, and I hope SE reconsiders. The point is precisely to do away with "hnq boogeyman" attitudes and actually get some useful data. It serves no one to bury our heads in the sand: (some) HNQs are harmful, and we need you guys to help us figure out which ones, how many, and how we can prevent it (and ultimately make HNQ better).
    – E.P.
    Commented Dec 6, 2016 at 13:10
  • 1
    I'm also disappointed by the lack of CM responses to Please open the Hot Network Questions to auditing via the Data Explorer.
    – E.P.
    Commented Dec 6, 2016 at 13:11
  • Here is one concrete example of a question raising eyebrows due to a high viewcount, which is likely not due to the HNQ effect - but thanks to the non-existent audit trail, we have no way to know. Whoop dee doo.
    – E.P.
    Commented Aug 11, 2018 at 11:19
  • Audit trail @E.P.
    – Shog9
    Commented Aug 11, 2018 at 17:25
  • 2
    @Shog9 Thanks for doing the digging - but this is the kind of analytics that ought to be internally available without relying on the (spotty) Wayback Machine coverage, particularly as regards determining past HNQ status. The performance reasons cited earlier really don't cut it in my view (i.e. if the data is too much, then sampling what you can store is still plenty valuable) and it is a shame that SE isn't investing any resources into fixing these sorts of problems. Hopefully the renewed attention to Q&A will bring more focus to these issues?
    – E.P.
    Commented Aug 11, 2018 at 17:49

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