8

This may be a duplicate of Should we stop auto-deleting dupe-closed questions? but the wording of the staff answer seems to indicate that it was effectively declined in 2012. It's now 8 years later and I believe that this should be revisited.

The full set of rules for automatic deletion is found here. I could not find a more recent set of rules.


On Software Engineering, this question (10k on Software Engineering required to view) was posted on April 1. It was closed on April 1 by the asker accepting the duplicate suggestion. It was automatically deleted on April 11 by Community as part of the automatic clean-up process under the rule RemoveAbandonedClosed.

The assumption that I am making is that duplicates are good signposts. That is, they help people find the answers that they are looking for. This includes people searching via external search engines, people searching a single site, and the recommendations that are displayed when people are asking a question. As such, having duplicates are not necessarily bad. I would suggest that any duplicate that is otherwise on-topic, well written, appropriately tagged, and is not an exact copy-and-paste duplicate of another question is a good thing as it can help promote the site, especially through external traffic.

Based on the defined rules, my understanding is that this question may trigger several automatic deletion triggers. After 30 days, it would be deleted if has a negative score and has no answers. Even if the net score was not negative, it may have still been deleted after 365 days if it had a score of 0 (or less than 1 if the asker's account is deleted), has no answers, or has a low (<= age of question in days * 1.5) view count.

I believe that duplicates should be exempt from these rules.

The particular question that I observed as automatically deleted has 1 up-vote and 2 down-votes, resulting in a net score of -1. However, it's not clear why this question was down-voted. In my opinion, being a duplicate alone is not a good reason for down-voting, yet there may be people in the community who disagree. We have limited control over the individual voting habits of people who have a sufficient reputation score. I believe that voting may be a good indicator of the ability of the post to serve as a threshold, but a low negative score should be insufficient to trigger the automatic deletion. I'm not sure what a good threshold would be, but I would recommend looking at a score of -3 with at least 5 voters. To enhance this, perhaps even look at the time frame - check the question so many days after posting or after closure to see the voting score.

The view count also shouldn't trigger the automatic deletion for duplicates. The point is not necessarily to get eyes on the question but to be an index. A low view question still shows up in searching and could show up in recommendations.

The lack of answers and comments also has no bearing on duplicates. I'd even argue that a duplicate without an answer is an even better signpost as it collates the answers on the canonical version of the question. Questions closed as a duplicate should be exempted as a rule.

In short, questions that are closed as a duplicate would preferably be never automatically deleted. However, if automatic deletion is necessary, the threshold for deletion should be raised to a much greater net negative score and not consider views, comments, or answers.


As gnat pointed out in the comments, this question was identified as being removed by RemoveAbandonedClosed. However, the documentation for this cleanup reason excludes questions closed as duplicates - the earliest that it should have been deleted was with the 30-day RemoveDeadQuestions script since it was negatively scored, unanswered, and not locked. One upvote would have pushed it to the 365-day RemoveAbandonedQuestions script, but it would have likely be eligible for that because it did not have a positive score, was unanswered, was not locked, didn't have comments, and wasn't on a meta site.

has a score of 0 or less, or a score of 1 and a deleted owner has no answers is not locked has view count <= the age of the question in days times 1.5 has 1 or 0 comments isn't on a meta site

3
  • 1
    applying 9-days rule of RemoveAbandonedClosed in the case you referred looks like a bug to me, its description explicitly prohibits this: "not closed as a duplicate"
    – gnat
    May 3, 2020 at 10:34
  • @gnat That seems to be the case. But I still think that the problem would be delayed. After 30 days, it would be eligible for RemoveDeadQuestions. After approximately 365 days, the question would be deleted by RemoveAbandonedQuestions. I'm not sure that the score (especially a very weak negative score) or low view count has any bearing on the question being a signpost. May 3, 2020 at 10:38
  • 1
    well I think that 365-days roomba properly attempts to keep "good signpost" duplicates by using low viewcount as a prerequisite for deletion. I am not sure though that the same viewcount should apply to duplicates as to "standalone" questions. Criteria for standalone questions seems to make sense, but if a dupe brings traffic to the target (by redirecting non-registered users) at a rate of say, one view a week / 5-6 views a month, maybe we better consider it good enough to keep. Maybe a separate cleanup RemoveLowViewsDuplicates would make better sense here
    – gnat
    May 4, 2020 at 15:12

3 Answers 3

5

In short, questions that are closed as a duplicate would preferably be never automatically deleted.

I fully agree with you that duplicates can be useful signposts. But putting up too many signs can drown out other useful information at times. Real life bilingual signposts are useful, but you don't put a second and third monolingual one in the same spot.

The same goes for duplicates, in my mind. A good duplicate is one that provides information/search terms that aren't already present somewhere else. Since not every duplicate does that, I think duplicates should sometimes be automatically deleted.

However, if automatic deletion is necessary, the threshold for deletion should be raised to a much greater net negative score and not consider views, comments, or answers.

I think I can agree with the part about comments and answers. Like you said, not having answers under duplicate questions makes them better signposts and in my opinion, having many comments doesn't automatically mean the question will make a great signpost.

I'm a bit more hesitant about votes and views.

The threshold you recommend (-3, 5 voters) might be too steep for smaller sites. Some sites are better at downvoting than others, and especially now that some sites can have the close-vote limit set so that there's only 3 votes needed to actually close a question, expecting 5 votes may be too much.

I'd much prefer it if someone would look at the data to see how many questions closed as duplicate with 1, 2, 3, etc. votes were deleted automatically, and what adjusting the threshold would mean, before recommending anything myself. How much stuff would stay undeleted then? How much of those are really 'unique' duplicates, do those really differ a lot from what's already still out there?

When it comes to the number of views, a signpost is basically useless if no one ever sees it. Having little or no views after being closed as a duplicate probably means that people manage to find their answers via other routes, perhaps other duplicates that turn up higher in the search results. At some point, there's going to be a lower limit to a signpost's usefulness. So I would not forever discard this, though perhaps the same questions as for votes may be good to ask: How much stuff would stay undeleted if this was removed, and how much of that is good?

All in all, looking into adjusting the thresholds seems useful to me, but what you're proposing here seems like it might leave out some thresholds that may be necessary and others that may be too high for smaller sites.

4
  • 5
    I'm sceptical that an automated process can fully distinguish between good dupe "signposts" & bad ones, but I think it's worth attempting. Views should be an important factor in that calculation. As I'm sure you know, when someone who's not logged in finds a dupe signpost via a search engine they get automatically redirected to the dupe target. I think that means that the signpost's view count doesn't get incremented, only the target's does, but I'm not certain.
    – PM 2Ring
    May 1, 2020 at 12:29
  • 2
    @PM2Ring Oh, that's a good one! I did not take that last part about redirections into account, but yeah. Would be nice to know if that counts as a view or not, and if not if there's perhaps another way to 'count' these.
    – Tinkeringbell Mod
    May 1, 2020 at 12:41
  • 1
    I think there are also ways to remove signposts. For example, hiding them from the homepage for logged-in users below a certain rep threshold. And I do think that we need a way to mark bad sign-posts. But a 0 or -1 post with 1, 2, or 3 votes probably isn't the right threshold. And there probably is some per-site configuration. I just have evidence of at least one bad dup deletion on Software Engineering - we need more data from across the network. May 1, 2020 at 13:47
  • 3
    I wanted to throw out an actual concrete proposal based on the experience and data that I have. I've found that concrete proposals are more effective than "we should think about this". At least with a concrete proposal, you have a starting point for designing and running an experiment. May 1, 2020 at 13:48
4

Most of my top search results from Duck Duck Go are pointing to closed "duplicate" questions. Many times those closed questions are higher quality, and have higher quality answers than the question that was kept. While I understand the need for housekeeping, I don't believe the process for marking duplicates is that effective at choosing the best question/answer set rather than the first question/answer set.

1
  • 1
    This is true, and I fully agree that manual curation is necessary to make sure that the best content is the target of duplicate closure. Still, if those questions have answers (and especially upvoted answers), I believe that would bypass the cleanup scripts that I'm referring to in the question. Typically, automatic deletion doesn't apply to answered questions. The concern is with signpost questions that, other than being a duplicate and being downvoted for that reason, are deleted automatically, which removes the signpost. May 3, 2020 at 10:33
0

Based on our current roadmap, we don't have immediate plans to work on this request. We'll revisit it at a later date and provide more details here when it has been prioritized and/or completed.

You must log in to answer this question.

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