There are several questions asking for a good way to model something. Are they in scope, or should they be closed?

Because a recent example is “Semantic distance between excerpts of text.” by John Berryman, I will take this as an example, but I would like to discuss a general issue.

Some people argue that this question is really about natural language processing and that without a rigorously defined model, the question is not about theoretical computer science and therefore is off-topic. I almost agree, but not completely, because it seems like a plausible possibility that questions like this can lead to interesting answers in the theoretical computer science.

I am not sure if I want to see more questions of this kind or not. On the one hand, it is interesting to see connections of theoretical computer science to various fields of computer science, and encouraging this kind of questions increases the chance of seeing such connections. On the other hand, if 20% of the questions on the top page are like this and few of them get answers interesting to me (and worse yet, they are bumped by community ♦ from time to time), I am sure that I will get sick of those questions.

Added: Another example of the how-to-model-this questions is “Finding islands of vertices in a network of roads containing one-way streets” by shindigo. This question was closed, and I was one of those who voted to close it. Honestly speaking, I did not realize that this question was in the same “how-to-model-this” category until Suresh pointed out in a comment to the question (silly me). I hope that comparing these two questions (and possibly other questions in the same category) may shed some light on what we want to aim at about how-to-model-this questions.


A huge part of our job description as theoretical computer scientists is finding formal ways to model informally-specified notions! (What is it that Turing did when he defined Turing machines in the first place?) For that reason, I don't think it's possible to define "modeling questions" as outside the scope of TCS, without more-or-less eviscerating the subject.

Having said that, there are better and worse ways to pose a modeling question. I think the best ways would make it clear (a) why TCS (as opposed to a different field, like systems or AI) might be well-equipped to answer the question, and (b) what a helpful answer would look like. (In other words: if the only answer to your question you could reasonably expect from a TCS expert is "I dunno," why are you asking it here? :) )

In the present case, I think the question is well within scope (and indeed, it did elicit a helpful, relevant answer from Shiva Kintali), but I might have phrased it like this:

"What numerical measures of similarity between natural-language sentences have been studied, and does the computation of those measures lead to any interesting algorithmic problems?"

  • 2
    $\begingroup$ Thank you for a very interesting view, and also for the concrete suggestions about how to pose a modeling question in a good way. I think that your first paragraph explains the source of my vague feeling that there may be something inherently wrong about closing this kind of questions just because it is not yet rigorously defined. $\endgroup$ – Tsuyoshi Ito Oct 15 '10 at 21:51

I am not, myself, a theoretical computer scientist, so I knew that I was at risk of being off topic on this site. However, it seems to me that one of the great benefits of a site like this is being able to inquire of one's peers about where to start looking for something. Interesting things happen when an individual jumps from a field that they are familiar with into a new and unfamiliar field. However, it's also at this point that the individual needs the most help. Google is certainly the first place to look, but after checking for obvious answers first, one can usually move a lot quicker if they just ask the community to point them in the right direction.

  • 3
    $\begingroup$ Hi John, it would also be helpful if you could guide the answers, to indicate what it is that you were looking for. For example there are at least a few answers that seem (to my eye) helpful, but you're the best judge of their utility. $\endgroup$ – Suresh Venkat Oct 15 '10 at 21:03
  • $\begingroup$ Thank you for explaining the situation. There may not be an ultimate distinction between theoretical computer scientists and “applied computer scientists,” but I assume that there is some distinction for now. I understand that it is very difficult to interact with people working in a field which you are not familiar with. However, in my opinion, the primary factor deciding whether a question is acceptable on the website or not should be whether it is useful/interesting to people working in TCS, not whether it (or its answer) is useful to people working in other fields. (more) $\endgroup$ – Tsuyoshi Ito Oct 15 '10 at 21:59
  • 1
    $\begingroup$ (cont’d) Therefore, I do not view “the question is useful to people working in an applied field” as a valid reason to keep the question open. I know that I am being selfish, but this is my honest personal opinion. It is not my primary purpose to make the website useful for people working in other fields. I can only hope that the efforts to make the website useful for the insiders will also make it useful for other people in the Adam Smith sense. (As I said, this is my personal opinion, and probably many people disagree.) $\endgroup$ – Tsuyoshi Ito Oct 15 '10 at 22:00
  • $\begingroup$ @Tsuyoshi I respectfully disagree with you. I think that having connections with external fields is a way to ensure that new ideas come into the community and good ideas from within the community move out to other communities. However, this being said, I also realize that there is also a limit to the amount of external connection the TCS exchange can have before the community becomes watered down and ineffective. However there are two elements in place to prevent this from happening: the TCS exchange FAQ's "What kind of questions can I ask here?" (needs work) and the power of the down vote. $\endgroup$ – John Berryman Oct 16 '10 at 2:38
  • $\begingroup$ Thank you for the reply. I know that interaction with external fields can be also beneficial to the community, and that does not challenge my point. My point is that questions do not become suitable on this website just because they are useful for people in applied fields. This does not mean that questions in applied fields are not suitable. If they are also interesting to people in TCS, I think that they are suitable. An FAQ entry may be good, but as you say, it needs some work. Downvoting is not an ideal tool to discourage certain kind of questions, although it can be used in that way. $\endgroup$ – Tsuyoshi Ito Oct 16 '10 at 22:05
  • 1
    $\begingroup$ I'm on John's side (also because I spend a lot of time talking to applied folks about their problems). It is a source of great frustration to me that folks in applied realms don't use what are (to me) obvious algorithms and ideas that would make their life a lot easier. But I recognize that there's a large barrier that they have to cross to see this, and I try to help ease that. While this site should mostly be focused on the needs of the community, I don't see a harm with quick and easy answers that help guide applied folks towards the relevant theory. $\endgroup$ – Suresh Venkat Oct 17 '10 at 17:07
  • $\begingroup$ @Suresh: I think that both the opinion of you and John and the opinion of me are reasonable, and the difference is entirely about what we want to aim at. I do not think that I have a strong opinion about the direction we should aim at, except that I wish to arrive at some consensus shared by many users of the website. $\endgroup$ – Tsuyoshi Ito Oct 17 '10 at 19:03
  • $\begingroup$ actually I don't think there's a lot of disagreement. my view is that the PRIMARY purpose is as an internal resource for the community, and a SECONDARY purpose is to assist with more application-oriented questions. Thus, as a secondary purpose, questions that model problems are welcome within reason and upto a point. $\endgroup$ – Suresh Venkat Oct 17 '10 at 19:34

I think for now, a conservative approach (leaving the questions as is) is fine. The fact that it got only two close votes reflects this, I think. I was one of the people skeptical of the value of the question, but I agree with your point that one of the features of TCS is that we can model lots of questions coming from more applied domains, and within reason it's ok to entertain such questions.

Should we have a 'modelling' tag for such questions ? or 'applications' or something like that ? so that people who don't care for such questions can ignore them ?

  • $\begingroup$ Thanks for an answer. It may be a good idea to create a tag for this (though I am not sure what the name of the tag should be), but we have to remember that there is another effect of creating a tag: the existence of a tag can be viewed as a sign that as a community, we accept questions of that topic and therefore it can effectively encourage questions of that topic. As I wrote, currently I am not sure if this kind of questions should be encouraged or not. $\endgroup$ – Tsuyoshi Ito Oct 15 '10 at 20:13
  • 1
    $\begingroup$ I like "modeling" more than "applications". (Another possible tag is "how-to-model".) $\endgroup$ – Kaveh Oct 15 '10 at 20:39

The one-way street question was far from research level and hence not on-topic according to current policy for that reason. If we think that question should be on-topic then we need to ask ourselves "should practitioners get an exemption from the minimum question level rules", not "is modeling ok".

I don't like the semantic distance question for a couple of reasons unrelated to modeling.

  1. The asker didn't attempt to teach us what they knew about the area so we could learn about it and/or answer their question better. This falls under the math overflow how-to-ask document's heading of "provide background and motivation" and "do your homework".
  2. I didn't think anyone in our community could give a good answer to that question without asking an AI person, and why not have the asker ask such an AI person themselves? It turns out other people (e.g. Shiva) knew more about that area of AI than I do, so this isn't the issue that I thought it was.

My vote is that well-asked modeling questions should be allowed, but perhaps we should wait until we receive some high-quality modeling questions some before deciding.

  • $\begingroup$ The one-way street question is “far from research level” in TCS, but the question is whether we should measure the difficulty level in our own terms. My opinion is yes we should (therefore it was right that the question was closed as too elementary), but Suresh has a different opinion. I agree that the semantic distance question was not asked in the optimal way. $\endgroup$ – Tsuyoshi Ito Oct 18 '10 at 10:43
  • 2
    $\begingroup$ Strongly connected components is something that most CS students learn as undergraduates. I think that's sufficient to qualify the question as not research level in any field. $\endgroup$ – Warren Schudy Oct 18 '10 at 13:25

I felt uneasy about the question, but after reading your and Suresh's posts, I agree. I was not sure if it was on-topic or off-topic, and therefore didn't voted to close (I also prefer if the question is closed by people who are more familiar with the topic.)

I think we should make the decision about closing "how-to-model" questions on case-by-case bases. As you have said, the question can lead to interesting TCS problems.

If I see a question that is asking for something that cannot be answered, will be argumentative, or very subjective, then I would vote to close it. To be able to model an informal problem, one probably needs lots of familiarity with the subject, and without feedback from the people working in the area it would be difficult to see if a suggested model is really capturing the intended informal question.

On the other hand, if the question is not too board/argumentative/subjective, if there are already studied models for the problem, such that the question can be answered, then I guess it is reasonable to leave it open.


I think the part of the problem is that we may end up with too many how-to-model questions. My two meta-conditions does not help with this type of question:

  • Is it OK to ask it from a college in a coffee break? Maybe, surely not too many of them.
  • Is it interesting to general theorist? Maybe, it depends.
  • $\begingroup$ (1) I share your uneasy feeling. If the question is closed as off-topic or subjective, I can understand the reason. (2) I agree that it is probably better to let people who are familiar with the topic decide whether the question is good or not, especially because it is sometimes difficult to tell whether the question is answerable or not if I do not know the topic. $\endgroup$ – Tsuyoshi Ito Oct 15 '10 at 20:56

You must log in to answer this question.

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