2
$\begingroup$

In a previous question I tried to provide information from data.StackExchange to answer a question about the math.SE

However, it may have missed the mark. I learned a high score doesn't necessarily mean a good question, just an accessible question.

What are the signatures of good questions on math.SE and is it possible to measure using data tools we have available such as http://data.stackexchange.com?

It's true we have all seen highly ranked questions that aren't quite as exciting as the numbers suggest. What do votes actually measure then?

EDIT I have fixed the title in response to all the tautological/troll comments below.

$\endgroup$
  • 6
    $\begingroup$ Votes measure how many users with the appropriate privilege have clicked on the arrow. Within the same page, vote count is correlated with quality, but across the site, the correlation is at best minuscule. I have seen average or below posts with higher vote counts than some good posts have view-counts. $\endgroup$ – Daniel Fischer Nov 7 '14 at 18:46
  • $\begingroup$ I don't know. Votes per view probably is correlated enough that it indicates something. But of course it's not reliable. $\endgroup$ – Daniel Fischer Nov 7 '14 at 19:02
  • 2
    $\begingroup$ @DanielFischer Votes per view correlate with "title is specific enough so that those not interested didn't open the question". $\endgroup$ – user147263 Nov 7 '14 at 19:04
  • 2
    $\begingroup$ @Rafflesiaarnoldii And that is part of a good question, a specific enough title. (And can you slow down the name-changing please?) $\endgroup$ – Daniel Fischer Nov 7 '14 at 19:07
  • $\begingroup$ Votes measure the difference between the number of times users have clicked an odd number of times on the "up arrow" and the number of times users clicked an odd number of times on the "down arrow" to the left of the post. $\endgroup$ – Asaf Karagila Nov 7 '14 at 20:50
  • $\begingroup$ (John: I don't know how many years you've been online, and where you've been to around the internet. But my comment is not a "troll comment", it might not be an extremely helpful one, but it's not an attempt for trolling. You should spend more time in dark corners on IRC or some other cesspools of teen angst to see what trolling is.) $\endgroup$ – Asaf Karagila Nov 7 '14 at 20:53
  • 3
    $\begingroup$ It seems to me that votes measure, more than anything, how relatable a question is. I don't know how to search the site, nor so I want to learn it; but from what I have seen, the highest vote counts tend to be for truly trivial questions, those seeking personal advice (extra points if the story involves personal hardship, true or imagined discrimination, and endurance), and - most popular of all - on stories glorifying mathematicians and so users themselves (anecdotes (veracity optional), why do you answer?, amazing achievements, ...). $\endgroup$ – gnometorule Nov 7 '14 at 20:56
  • $\begingroup$ @Asaf are you sure about that? Just clicked UUUD, gives -1. (I did not, but it would.) $\endgroup$ – quid Nov 8 '14 at 10:58
3
$\begingroup$

I'll follow up on daOnlyBG's suggestion

If you want to see some great problems, look for the ones with high star ratings.

As Arthur Fischer pointed out, the leaders in total number of favorites are mostly soft questions. I wrote a more elaborate query: Questions favorited by a user with a matching tag badge which does a better job of identifying challenging problems. The top results are skewed toward hard integrals: I think this is partly because so many users have badge, and partly because integrators tend to favorite questions more actively.

(As a byproduct of how the query is written, a user counts more than once if they have more than one matching tag badge, but this seems reasonable.)

There is a similar query for unanswered questions only, although due to Data Explorer being up to a week out of date, some of these have been answered meanwhile.


By the way, Stack Exchange is currently looking for a data science engineer:

We're building our data team and are looking for a data scientist with strong programming skills, or a developer with strong data science inclinations. Some problems you could work on:

  • Identify low-quality questions on Stack Overflow
  • Recommend questions I can answer, or might be interested in reading

If you succeed in identifying good questions based on Data Explorer, they might hire you...

I think identifying good questions is harder than identifying bad questions. It's somehow the opposite of the Anna Karenina principle: bad questions are all alike, every good question is good in its own way.

$\endgroup$
1
$\begingroup$

I know this might incite some eye-rolling, but "good" is a subjective term, even in this context. For all I know, a good "math" question could be one with symmetry in either the problem or the solution, or some clever trick that cancels out a bunch of terms (i.e., telescoping sums). Nevertheless, I'll assume that by "good" you mean "thought-provoking."

On this site, I would not trust the "vote" parameter much, for two reasons. First of all, people can vote on a problem they mildly feel is worthwhile, and forget about that particular problem soon thereafter. That action isn't reflective of a very thought-provoking problem. Furthermore, while a "good" problem might indeed draw many votes, it doesn't have to, either. There might be some correlation, but in my humble opinion, it's not a very strong one.

Second of all, there is an incentive for those who like to earn badges on this site to "vote up" as many posts as possible. Sure, MSE tries to hinder this via a cap on how many votes you can do per day- but that doesn't stop some users from just starring random questions, especially if it means getting closer to whatever badge they want.

There is another parameter on StackExchange that I like very much- one that addresses both of these concerns: the "favorite question" star. No one gets any award for starring as many posts as possible, so it's not like there's an incentive to loosely hand them out. Furthermore, starring a question allows you to keep track of it, so on your home page, you're always getting notifications for the development/conversation for that problem and its respective solution.

If you want to see some great problems, look for the ones with high star ratings. And check out the high-vote, high-starred unanswered questions as well. Many of them are open, and very thought-provoking.

TL;DR The favorite-question star is an indicative method of finding "good" questions.

$\endgroup$
  • $\begingroup$ IMHO, this list seems pretty full of banal soft questions. $\endgroup$ – user642796 Nov 7 '14 at 22:20
  • $\begingroup$ on meta.math.SE users complain about the low quality of questions, but since you are attacking my use of the word "good" what does that even mean? $\endgroup$ – cactus314 Nov 7 '14 at 22:27
  • $\begingroup$ I'm not "attacking" anything. $\endgroup$ – daOnlyBG Nov 7 '14 at 22:41

You must log in to answer this question.

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