14

So as part of my mucking about with the StackExchange API I decided to check what kind of questions high-rep people tend to ask.

I pulled in all the question data and then took the rep of the user asking each question and grouped those by the question tags:

$reputationTags =
  GroupBy[
   Flatten@
    Normal@
     Map[
      Thread[#owner["reputation"] -> #tags] &,
      $questions[All, {"tags", "owner"}][
       Select[#owner["user_type"] != "does_not_exist" &]
       ]
      ],
   Last -> First
   ];

Then taking only those tags that get a lot of love on the site (i.e which are used more than 100 times (in this case) I find something interesting:

repTags[n_: 0] :=

  Reverse@Sort[
     N@*Mean /@ Select[$reputationTags, GreaterThan[n]@*Length]] // 
   Dataset;
repTags[100]

tag usage

Most of these tags make sense to me--they're poorly documented or new or require a certain amount of knowledge to make sense of--until we get to "output-formatting". Note that this is higher even than "graphics".

So what makes "output-formatting" tricky? I'm guessing there are things to learn here which I do not yet know.

  • 2
    Without digging through the tag to refresh my memory I would guess that these questions relate to formatting output in a way that isn't natural or obvious, or modification of existing output in specific ways. I further guess that the reason that "high rep" people ask these questions is because they are about Mathematica itself rather than applications of Mathematica, and that "high rep" users tend to be interested in the software itself or they wouldn't be as active as they are here. It may also serve as an "umbrella term" for questions that are otherwise difficult to tag. – Mr.Wizard Jun 15 '17 at 22:27
  • @Mr.Wizard I'm currently digging through and comparing "graphics" and "output-formatting" and that does seem to be the case. That is, even though you and Szabolcs have asked many more "graphics" questions, there are enough people for whom their "graphics" question is their only interaction on the site to bring the Mean down. Perhaps the Median would have been a more interesting statistical criterion here. – b3m2a1 Jun 15 '17 at 22:30
13

You get a different view of things if you examine the number of questions asked by high-rep users. For instance, for users with rep > 2000:

ClearAll[repTags];
repTags[nQ_: 0, minR_: 0] := 
  Reverse@Sort[
     Length@*Select[GreaterThan[minR]] /@ 
      Select[$reputationTags, GreaterThan[nQ]@*Length]] // Dataset;

repTags[100, 2000]

Mathematica graphics

is ranked 28th (38 questions). So while the average rep of the askers of output-formatting questions is high, the number of questions asked is low, while the number of questions is quite high. rises to 13th (22 questions) if the rep limit is set to 10000, which is in line with the high mean rep of the tag.

There's a similar drop in the ranking of . My guess is that , like , is difficult to accomplish sophisticated things, and high-rep users are more likely to be interested in applications for other end-users and other things that call for sophistication. Mr. Wizard's comment that high-rep users are interested in Mathematica itself also seems to me to be a likely reason (e.g., Rojo's pacman drinking beer question). The mean-rankings depend on the relative proportion of high to low rep. Many of the low to medium rep users seem to be interested in M as a tool and are principally interested in getting the right answer, not in (unless you include as a special case of output formatting, but I am thinking of a plot as a sort of answer or solution).

[And it's nice to see ahead of in this list, because it seems much harder to get things just right with plotting than with graphics.]


Interesting transitions in ranking depending on rep (10-1000, 7500-8000 -- if you go higher, the number of questions gets small enough that the difference of a couple of questions can make a great change in ranking):

Mathematica graphics Mathematica graphics

With[{keys = Normal@Keys@repTags[140, 1]},
 ListPlot[MapThread[
   Callout[Tooltip[#1, #2], #2] &,
   {Transpose@
     Table[Thread@{k + 1, 
        34 - Ordering@Ordering@Normal@Values@repTags[140, k][keys]}, {k, 
       Range[1000, 10000, 500]}],
    keys}],
  Joined -> True, ImageSize -> {Automatic, 400}, 
  ScalingFunctions -> "Reverse", Frame -> True, AspectRatio -> 1., 
  PlotRange -> {{1000, 10001}, All}]
 ]
  • Nice analysis. Good apply a chop on the min rep to drop the lower outliers. Also, just looking at your Dataset it's kinda cool that "neural-networks" is so popular already as, unless I'm much mistaken, they debuted in v11. – b3m2a1 Jun 16 '17 at 15:20
  • Indeed, fun analysis. Thanks! – Mr.Wizard Jun 16 '17 at 18:02

You must log in to answer this question.

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