30

The required view count to avoid auto-deletion of a question when score = 0 is too high.

As an example: http://superuser.com/q/1472443/116475 had 562 views within 1.1 year. This is way more views than average for the site (1 or 2 standard deviations or so), but yet the question was still removed because it didn't have enough views. Currently the rule is "has view count <= the age of the question in days times 1.5."

This threshold makes sense for Stack Overflow, where due to its largeness, people frequent questions a lot more, but doesn't make sense for smaller sites that have lower traffic proportions.

The required view count could instead depend on website-specific view count stats to account for the fact that some SE websites are much less frequented than others (e.g., remove if count is in the lower quartile on the given SE website, or better use human judgments on some question subset to decide).


Example of incorrect deletions caused by Roomba on Android.SE on my account only over the past 60 days (since I joined SE, I've had hundreds of such questions removed):

enter image description here

or on Superuser:

enter image description here

or on WebApps:

enter image description here

Five questions (with zero score) posted by just one user within 60 days were removed, despite most of them receiving an average number of views for the site, just because the overall average views per question on the site is lower than on larger sites.

or on https://softwarerecs.stackexchange.com/:

enter image description here


The following table shows that the median number of views on a question on WebApps is only 42 after 12 months (source, 25%, 50% and 75 are percentiles), which is 13 times less than the Roomba minimum number of views to avoid automated deletion:

enter image description here

Only the top ~10% of the most viewed questions posted on WebApps have more than the Roomba minimum number of views to avoid automated deletion. This means that in practice, if a question has a score of 0 and no answer and less than 2 comments, it'll get deleted.

22
  • 9
    We could just scale the factor based on total site views. I bet this 1,5 was based on some historical amount of site visits on SO. It could be time to revise this completely, as today chances are that siteviews on SO are higher as well, and the factor 1,5 might be too low for SO, whilst it is too high for smaller sites.
    – Luuklag
    Oct 7 '20 at 19:45
  • 2
    True, Community Building does get some questions with few views over several months; yet the site is a valuable addition to our list of sites. I checked their meta and couldn't see anything about having a different formula for the Roomba, maybe it's turned off and deletions are done manually.
    – Rob
    Oct 7 '20 at 20:30
  • 1
    Franck's similar question; tuning the algorithm based on site, but for child metas.
    – Rob
    Oct 7 '20 at 20:39
  • 1
    Another note: a score of 0 doesn't necessarily mean that no one values the question because they never bothered to vote on it. It could also mean that one or more people thought the question was useful, but an equal number of people disagreed with the question or what it was asking about (rather than think the question was of low quality or not useful). Since at least some people thought the question was useful, I don't think such questions should be deleted. Oct 27 '20 at 2:30
  • 3
    I feel like your new screenshot kind of undermines your point. As you say at the top, that question had 562 views on becoming 1.1 years old, which is "1-2" standard deviations from the average views for a question, but the questions listed there have very little views compared to that. I'd list some examples of questions that have a couple hundred views or more, along with some traffic statistics for the site. (As you know, I understand the request and am in favor of implementing it; this is just some advice on how to make a stronger case for the request.) Nov 22 '20 at 20:32
  • 2
    Again, your new (third) screenshot still has the same issue as in the above comment: apparently very little views. If questions on that site actually had that little views on average, it would be nice if you showed some traffic statistics to indicate that that was the case. Nov 23 '20 at 20:44
  • 6
    As with many requests like this, you've just assumed the deleted questions were valuable to the site, that the site would be better with these questions not deleted, and that the Roomba "shouldn't" have deleted the questions. All of the examples for which you've provided merely screenshots of the titles are your own questions, which means your evaluation of their quality is far from unbiased. I'm not saying they should've been deleted or weren't good. I really have no idea of those questions' quality. But, you need to at least demonstrate the underlying assumptions have some validity.
    – Makyen
    Nov 29 '20 at 7:22
  • 4
    @Makyen how can I access questions that aren't mine and were deleted by Roomba? Nov 29 '20 at 7:23
  • 2
    Here is the reason for roomba: "Abandoned, unanswered questions can be a nuisance for readers when they appear in search results. While every question deserves a chance to be answered, at some point the annoyance to those searching for a solution outweighs the increasingly-small chance that an answer will be provided." I wonder how are "annoyance" and "chance to get an answer" measured, and how both of them affect the growth of a site (or more importantly, the mature growth of a beta site)
    – Ooker
    Dec 11 '20 at 5:35
  • 1
    I think (though I'm not sure) that the search result they mention is the site's search function, not Google. Because only in there the user's experience can be reduced?
    – Ooker
    Dec 11 '20 at 5:43
  • 1
    @Ooker I see. Then roomba should only keep questions with few views then, since few views mean that it bothers nobody. But really, more seriously, what a lack of respect to the OPs to remove the content when it is perfectly valid. Dec 11 '20 at 5:45
  • 2
    But I wonder, if a question with a lot of views cannot get even a vote, does that mean it's really uninteresting to many people?
    – Ooker
    Dec 11 '20 at 6:38
  • 2
    So that question had "562 views within 1.1 year", which "is way more views than average for the site", and none of those viewers thought it was useful content? It's hard to see why a change should be made to keep that around. Or was the situation, as @Sonic suggests, that it had some upvotes but an equal number of downvotes?
    – jonrsharpe
    Dec 11 '20 at 10:15
  • 5
    @jonrsharpe The code, as written, still unfairly biases SO and other large sites with much more traffic. Why should questions on such sites get "special" treatment compared to questions on lower-traffic sites? If you believe that such questions should still be deleted, then all sites should be treated equally in that regard, which means that the required view threshold to prevent deletion should be higher on high-traffic sites. At least some form of traffic scaling should be implemented, even if not in the exact form as proposed in this question. Dec 11 '20 at 10:25

You must log in to answer this question.

Browse other questions tagged .