-11

Are the different SE sites intended to operate as a whole? Or is there no expectation that any one site should care about getting questions answered unless they are on topic for that site?

Use case: Someone who is just starting to get into a topic and hasn't asked questions about it before is trying to figure something out and needs to choose where to post the question. How do they figure that out?

Give the number of different SE's, many with close (or even even overlapping) topics, it seems that figuring out what SE to post a given question to, particularly when the poster is the kind of person who could most benefit from SE (someone who is new to the domain) is a non-trivial problem. This is made worse by two thing:

  • The front pages doesn't have an easy-to-find link to the "what's on topic" page, adding at least a few extra click to find that (and that's assuming you already figured out you have to actually start asking a question to find that link).
  • The normal process for reacting to an off-topic question (closing it with a boilerplate comment), while more polite, isn't functionally much more helpful for getting the question answered than being told "STFU and go away".

Has any effort been put into this user story?


A few features come to mind that might help:

  • Including a "consider posting to X" option in the closes-as-off-topic action. Maybe even going so far as to requiring the user to pick something (with "none" being an option) to close a question.
  • Has anyone tried throwing ML at the problem of "what site would like this question"? Having the new question page able to suggest a better site (and an easy action for "copy what I've already typed") could be an effective way to avoid the problem in the first place.
33
  • 6
    Each community is it's own, and the rules and guidelines for each community vary wildly given it's mostly defined by the community itself. It's unreasonable to expect someone on Stackoverflow to know what is and isn't on topic on devops, and thus it's irresponsible to suggest someone should ask on another site without knowing well what is on topic there.
    – Kevin B
    Feb 19 at 17:56
  • 8
    By going to the help center for the site they wish to participate in.
    – Kevin B
    Feb 19 at 17:58
  • 1
    We tried the "consider posting on X" option. People just put anything in, flooding less popular sites with poor quality questions so that option is a non-starter. It would only work if you were limited in some way to suggesting sites you're active on e.g. the list of suggested sites was those you can close/reopen vote on but that would make the list zero for most people. Feb 19 at 18:00
  • 5
    "The front pages doesn't have an easy to find link to the "what's on topic"" the tour is linked on every page: in the footer and in the (?) icon at the top right. If you click on "Ask question" you even get a quick description on the right for what the site is about. And a list of helpful links. It blows my mind that so many people come here and say the information is missing. Sure, you didn't find it but that doesn't make it not-there. Moreover, the information is not only where I'd expect it to be as a new user it's even more accessible.
    – VLAZ
    Feb 19 at 18:03
  • 3
    Like, I'm used to visiting a website and having to search for their help centre or equivalent to figure out how to use the site. On SE there are multiple links to it on the most usual places where one would look. And the information is presented when you post. I've not seen a website which is that generous with the usage information.
    – VLAZ
    Feb 19 at 18:04
  • 1
    @VLAZ If so many people say it's missing, then they are correct. There is no meaningful difference between "not there" and "people don't find it". I suspect that issue would vanish by just adding a "what's on-topic" link (using that exact text) to the front page.
    – BCS
    Feb 19 at 18:05
  • 3
    @BCS and when people overlook that one, then what? Use mental control to beam the information directly into their brains?
    – VLAZ
    Feb 19 at 18:09
  • 2
    We provide a lot of information but nobody reads it. They treat it as something that slows them down from asking the question that they really want to ask right now and nothing will stop most of them from doing that. Feb 19 at 18:12
  • 1
    They only care when they are banned. Feb 19 at 18:15
  • 7
    @BCS I believe previous experiments have shown that people will ignore any helpful text regardless of where it is located in their rush to ask their question. Feb 19 at 18:17
  • 3
    @BCS it's currently placed it a giant popup when you first open the ask wizard, on the right side of the ask wizard page 1, on the right side of the ask wizard page 2, in a banner when you sign up, and when your question is closed you are also linked to it. Where else do you think it should show up?
    – Kevin B
    Feb 19 at 18:17
  • 1
    That actually already happens, lol
    – Kevin B
    Feb 19 at 18:19
  • 2
    @BCS no such AI exists that can accurately determine a given question should be asked on $OTHER_SITE. We can't even create accurate recommendations currently for duplicates or similar questions on the same site.
    – Kevin B
    Feb 19 at 18:23
  • 2
    Yes, it's likely possible a solution could exist that would be better at recommending users what site to ask on than someone who doesn't know what stackexchange is. That's a fairly low bar, so low in fact it'd likely still be wrong most of the time.
    – Kevin B
    Feb 19 at 18:41
  • 7
    I'm just imagining the hostile meta posts "But the site said the question belonged here!"
    – Kevin B
    Feb 19 at 20:21

3 Answers 3

8

Someone who is just starting to get into a topic and hasn't asked questions about it before is trying to figure something out and needs to choose where to post the question. How do they figure that out?

  1. Identify the field of expertise that answers would lie within.
  2. Find a Stack Exchange site that matches that field.
  3. Follow the instructions in How to Ask.
  4. Post the question.

I'm more familiar with the Stack Exchange network, so I use a different approach:

  • List all sites the question would be on-topic on (off the top of my head).
    • When I can't quite remember a site's scope, I check its /help/on-topic page, and do a quick search through its meta.
  • Think about what sort of answers I want.
  • Choose the site that's most likely to give the perspective I'm after.
  • Post the question.

I find this a bit quicker, even though the order is somewhat backwards. Heuristic filtering should start with the cheapest heuristic: there are thousands of fields of expertise, but fewer than 200 Stack Exchange sites.

The front pages doesn't have an easy to find link to the "what's on topic" page adding at least a few extra click to find that (and that'a assuming you already figured out you have to actually start asking a question to find that link).

Iirc they used to, circa 2014. I find this page by visiting the site's homepage then adding /help/on-topic to the end of the URL.

Has any effort been put into this user story?

Lots.

Has anyone tried throwing ML at the problem of "what site would like this question"?

Is a question about Inkscape best on Stack Overflow, Super User, Ask Different, Unix & Linux, Ask Ubuntu, Graphic Design, Web Applications, Webmasters, User Experience, 3D Printing, Game Development, Wordpress Development, or Reverse Engineering?

It'd have to be pretty clever ML. Just looking at the words isn't enough, and really taking into account the order of the words is something even GPT4 and pals aren't all that great at. It might only take a few months to put together an expert system, but that wouldn't remain valid as site scopes drifted over time… I don't think this is tractable.

15
  • #2: "find a site". How they are expected to do that step is literally my question. #3 "How to ask" where on the home page for an SE is that link? Checking a random SE site and that text isn't there. Nor is the word "topic".
    – BCS
    Feb 19 at 18:30
  • @BCS go to: stackexchange.com/sites
    – rene
    Feb 19 at 18:32
  • @rene this is also, linked at the end of the tour.
    – VLAZ
    Feb 19 at 18:33
  • As for ML, 100% it couldn't be a static system. It would have to be re-tried on current data on a regular schedule. Luckily, the data curation (usually the $$$ step) is already done for you and a system that generates 3-5 "maybes" would be totally viable. It doesn't need perfect results.
    – BCS
    Feb 19 at 18:33
  • 1
    An option if you really have no idea where to start: SE search isn't great but with some keywords you might hit the jackpot: stackexchange.com/search?q=Inkscape
    – rene
    Feb 19 at 18:34
  • @rene "paradox of choice". Also, why isn't that at stackexchange.com/? The list of questions on the network home page seems more like a window display than something people would uses. Sure it's cool, but it's not the highest utility.
    – BCS
    Feb 19 at 18:36
  • @BCS What would be the input to such an ML system? Keywords straightforwardly don't work. I think some sites have a phrase-detection expert system, but that's a very coarse first-pass. If you can design an ML system that works, I think you could argue for at least an honorary doctorate.
    – wizzwizz4
    Feb 19 at 18:36
  • @BCS If you want fewer sites to choose from, you can select a category at the top – or in the footer of every page. (Scroll down from here, and you'll see them!)
    – wizzwizz4
    Feb 19 at 18:37
  • @BCS The list of sites is in the header of stackexchange.com/ in very readable text
    – Esther
    Feb 19 at 18:38
  • @wizzwizz4 the data set would be the full text and status of every question, answer, comment and chat in the DB across all sites. Run that thought an off-the-shelf embedding function and uses the per-site votes/answers/close state as the output.
    – BCS
    Feb 19 at 18:44
  • @Esther, Sure, but why is the current front page more useful than having that as the front page directly?
    – BCS
    Feb 19 at 18:45
  • 3
    the current front page on SO is useful for exactly nothing, and the community has no power to change it. I think a better home page that acts to introduce users to the community (for every stack including SO) would be quite worthy of dev time, but you'd have to convince SE of that, not us.
    – Kevin B
    Feb 19 at 18:47
  • 1
    @KevinB "People here" != "people who have ability to decide how SE work" might be the most important statement in this entire interaction. -- I was under the impression that this site in the intended way for user to interact with that second group. :-( That not being the case would explain many frustrations over the years.
    – BCS
    Feb 19 at 18:51
  • @BCS The algorithm you describe wouldn't work (too many confounders). But let's assume we have a perfect dataset, with a "degree of on-topicness" score for every site: the hard part is the embedding function. I've explained why (I think) an off-the-shelf embedding function wouldn't work well; do you have a particular reason to believe otherwise?
    – wizzwizz4
    Feb 19 at 18:59
  • @wizzwizz4 the technique for embedding you described in original answer (some kind of n-hot vector?) sound more like roll-your-own embedding functions that what I'm thinking of. I haven't done much direct work with ML, but last time I looked at things it took me all of 5 minutes to find OSS text embedding solutions that (claim to) include the kind of structure/semantics processing you say would be needed. If you have a budget, I'd guess much better is easy to get. I've worked with people who had access to what would be needed close to 10 years ago.
    – BCS
    Feb 19 at 19:10
8

What is the expected way for a NEW user to identify the correct SE for a question?

This is, in some sense, the wrong framing. Stack Exchange is not fundamentally about the flow "I want to know something -> post a question -> someone else answers -> I read the answer and walk away". That's how traditional discussion forums work. Stack Overflow was originally created explicitly to get away from that model for programming questions. (Later, Stack Exchange was created to build on that success and apply the same model to other subject areas.)

Every network site's tour includes a blurb along the same lines: "With your help, we're working together to build a library of detailed answers to every question about [things that are on topic for the site]." (Stack Overflow's tour also says that the answers are "high-quality".) The important idea here is that asking a question helps build the library, by: identifying an important topic that can be of use to others and which is not yet addressed in the library, and providing a place for the answer(s).

So instead, the right way to think about it, as a new user, is to identify what questions you have that are correct for the current SE. There might not be any, and that's fine. If you're interested in the topic, though, hopefully you'll at least be able to find existing questions - and their answers - that are helpful to you personally.

Granted, not every site really works like that. It turns out that you can't really force entire communities to use software in the specific way it was intended to be used.

Are the different SE sites intended to operate as a whole?

No. The selection of topics covered is pretty much completely arbitrary - it depends on what gets through the Area 51 process. While it's true that a few sites were initially created to offload not-quite-programming questions from Stack Overflow into more specific categories, this is more or less a historical accident now.

It seems that figuring out what SE to post a given question to, particularly when the poster is the kind of person who could most benefit from SE (someone who is new to the domain) is a non-trivial problem.

I have three points to make here.

  1. Someone who is new to the domain should not expect to know what to ask.

    I have experience being a beginner at multiple things, and I have looked back and reflected and noticed that common pattern. I could give examples but this would turn into a blog post.

    Beginners benefit from the fact that SE sites create a library of useful Q&A. Ideally, it's in the form of questions that can be easily found with a search engine. But since a beginner might not know what to search for (and sometimes the terminology is so heavily overloaded that the search terms overlap and find completely unrelated problems), we allow everyone to ask anyway - and redirect them to duplicates once we've figured out what they're trying to ask.

  2. In non-technical fields there is usually really not that much overlap between sites. Like, I'm sitting here trying to imagine a beginner wondering, "should I ask my question on photography.SE or X.SE?" and trying to fill in a value for X that could make it even remotely plausible that someone would actually wonder about this.

  3. If there is a real source of confusion - or if you just don't see an appropriate site in the list and want to double-check - then yes, there is a process. The process is: you come right here to meta.SE, check for previous meta.SE questions that look like they might be helpful, and finally ask. Present the general form of the question you want to ask - use block quotes to make it clear that you aren't actually trying to ask it here - and ask where it would fit.

The front pages doesn't have an easy-to-find link to the "what's on topic" page

Re-organizing the help to make this clear could be a separate feature request. But keep in mind that there is a lot of documentation that needs to be presented for each site. It simply won't be possible to put all of it in an immediately accessible place, without hindering access to... the actual site.

That said, if you go to the main Help page on any site, it will immediately present a table of contents, neatly organized into sections; the first section is titled "Asking"; and the first three entries there - marked as "pinned", which I assume means they can't be customized per-site - are titled "What types of questions should I avoid asking?", "What topics can I ask about here?" and "What does it mean if a question is "closed"?".

That seems pretty clear to me.

The normal process for reacting to an off-topic question (closing it with a boilerplate comment), while more polite, isn't functionally much more helpful for getting the question answered than being told "STFU and go away".

Well, yes. No site is responsible for sending questions elsewhere. The migration options are intentionally limited, because each other site has its own standards. This was especially problematic in the past for the Code Review SE, because people would try to dump questions there from Stack Overflow that aren't on topic for Stack Overflow, simply because the question already involves working code. That isn't nearly enough to make a valid question on Code Review.

Has anyone tried throwing ML at the problem of "what site would like this question"?

It wouldn't surprise me. But ML probably wouldn't do any better at that task than it would at attempting a direct answer to the question.

If people want to use AI, off-site, effectively as an advanced search engine (that rewrites content that it slurped from Stack Exchange sites, and occasionally makes up nonsense, and never sanity-checks your intended goal unless it looks like you intend to harm others), that's fine with me. As far as I can imagine, it would be fine with just about anyone else on the network, and it's not as if we could prevent it anyway.

6
  • FWIW: my experience with most SE larger sites (at least anything more recent than about 10 years ago) is that they have already accomplished (as in finished, completed, nothing constructive left to do) the "create a library" goal. It's literately been years since I've had a question I wanted an answer to where I haven't either found a preexisting answer to, or been told it's off-topic. IMHO, what would be more valuable than a carefully curated and highly exclusive library, would be a place people can ask a good questions, some of which get promoted into that curated library.
    – BCS
    Feb 21 at 0:33
  • As for ML: Informally, it produces output that's at best the median quality of it's inputs. Getting enough quality input to get quality output for general questions is unlikely to happen. -- But the data set of voting history on questions and answers, and closed questions is likely to be high quality. Being able to give the "odd this draft will get a good answer vs be closed" seems tractable. And once it's available for a bunch of sites, giving that metric on a draft across multiple sites is just a matter of paying for he compute.
    – BCS
    Feb 21 at 0:40
  • 1
    @BCS, It sounds like you want something other than what Stack Exchange provides. That's fine. There are many other sites other than SE where you can go; or you can try out SE Discussions to see whether it meets your needs. But regardless, that is a different topic than what your question above raises. I recommend keeping comments here focused on the proposal you articulated in your question.
    – D.W.
    Feb 21 at 2:02
  • 1
    I think you are overestimating what AI can achieve. I am not at all confident that it is feasible to build a bot that will accurately predict whether a draft of a question will be closed or not. I suspect there will be a high error rate, and require non-trivial effort.
    – D.W.
    Feb 21 at 2:03
  • @D.W. How to direct user to SE Discussions (if that's the right action); 100% in scope of what I'm asking/proposing. The SE network calling people expecting it to be something else a "them problem" and doesn't need to be addressed and then moaning that people keep "misusing" SE seem to be self contradictory.
    – BCS
    Feb 21 at 18:44
  • As for ML and "non-trivial effort", 100%. As for error rates; IIRC there's a proof that error rates are limited only by the amount of training data and how well defined "error" is. I expect the 2nd is the limit here, not the ML tech. I'd not be surprised if ML can more accurately predict consensus than any one moderator.
    – BCS
    Feb 21 at 18:44
3

It would be great to have a better solution, if we could find one. But it is not an easy problem to solve. I'm not confident that the changes you've proposed would have a positive effect. And at this point Stack Exchange, the company, devotes very little developer time to making improvements to the platform, so realistically any improvements need to make a big positive difference to have any decent chance of being adopted.

I wonder if there might be a misconception here. I wonder if you are imagining a user with a question, assuming that there must be some site where it fits, and so thinking that the goal should be to figure out which site it should be posted on.

But not every question has a home somewhere on Stack Exchange. In fact, I would suggest that most questions are not suitable anywhere on Stack Exchange. Most questions shouldn't be posted anywhere on Stack Exchange, at least not in the form that someone new to the site might naively write them down.

Moreover, most sites have their own quality standards and expectations. We don't want a user typing in a question, getting an AI suggestion to "consider site X", then the user posting it on site X without reading the help pages or learning about site X's expectations. (Unfortunately, there is an all too likely progression for how that might go from there: community members respond poorly to a question that doesn't meet the site's standards, the user complains "but your website said to post it here!", other community members respond negatively to a failure to respect site norms, the user has a bad experience, and leaves feeling that Stack Exchange is toxic.) Unfortunately, this unhappy pattern already happens with users trying to be helpful and suggesting other sites -- automating the process would probably not be helpful.

Instead, to participate in a fruitful way, new users will probably need to learn more about the Stack Exchange model. They'll need to do some searching to find the right site, do some research about norms and expectations on that site, and do substantial research to prepare a good question that fits well with the site scope. I don't think there is any way around that, and I don't think AI can change the fact that posting a well-received question on Stack Exchange often takes some work for a first-timer.

5
  • I think the feedback loop you describe already exists. While the SE networks is a good source of answers, I actively avoid anything that would count as "contributing" (not that I want to avoid that) because experience has demonstrated that questions that aren't already answered are unwelcome and will just be closed. IMHO this "actively discouraging everyone from asking new question" is likely the closest think to an existential threat to SE.
    – BCS
    Feb 21 at 0:46
  • BTW: by "first-timer" I'm mostly thinking of people who haven't used the specific site in question. That would include people who a so active on other sites that they haven't needed to look at the on/off-topic page in years. However they may be entirely unfamiliar with the rest of the sites. -- Maybe one option for the "but the bot said" issue would be to only list sites with high confidence classifications. Or go to the other extreme: Give an alphabetic list of the top 10 with no suggestion as to which is best.
    – BCS
    Feb 21 at 0:53
  • @BCS, Your concerns about answering questions seems like a separate topic, unrelated to your meta question here. I encourage you to keep comments here focused on the specific proposal you raised, rather than bringing in controversial unrelated matters. I don't know how experienced you are with AI, but unfortunately, current methods are not very good at assessing confidence (they are often confident but wrong). In any case, I encourage you to take the feedback, and think some more and see if you want to come back with a specific proposal along with evidence and data to support it.
    – D.W.
    Feb 21 at 2:01
  • See above, but tl;dr; the "specific proposal" isn't the ML etc. (those are random ideas that might help at the specific proposal). The "specific proposal" is "how to avoid disappointing and offending people when the majority of people who see SE sites are primarily motivating by answering a question?" (Preferably with having to ask a new one.) I think the architecture of the site is build to support the mods and people who answer lots of questions, i.e. one of the smaller minority populations. I'm proposing that that choice should be examined and either made explicit or abandoned.
    – BCS
    Feb 21 at 18:52
  • @BCS, OK, but that's not the question that is stated or asked in the post at the top of this page. The question at the top of this page is "Has any effort been put into this user story?" followed by some suggestions of features. If "how to avoid disappointing..." is the question you want to ask, I suggest you take the feedback you've received here, reflect on it, and then think about whether there is a new, separate question you can ask that will be fruitful.
    – D.W.
    Feb 21 at 19:02

You must log in to answer this question.

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