6

I originally wrote this as a comment on the State of the Beta question, but decided to move it into its own thread once it got pretty long and detailed.

I have what I think is a problem, with no particular solutions in mind: tags are not very useful right now, compared to the other SE sites I use.

The core problem, as I see it, is that it's not clear what tags are for. Some are general topics like . Others describe a relevant core feature like . Others describe an action you're taking, like . Others categorize by Emacs package () or external tool (). Then we have some special ones, like [emacs-24.4] for version-specific questions.

The problem is that these categories behave differently—often orthogonally. And the result is that we have a very long tail of tags with very few questions (5-15), which are mostly not very useful. Moreover, since there are so many different types of tags which can all apply to a single question, I suspect many questions are missing relevant tags, leading to even more fragmentation.

This makes tags more difficult to use for both asking and answering questions.

For asking, it's difficult to figure out which tags are necessary. Unless you're in one of a few very clear-cut cases (), you have to guess something abstract—and since so many tags have a similarly small number of questions, it's hard to naturally tend towards the same tag other people are using. As it stands, there is no real pressure to consolidate on given tags, so the fragmentation problem isn't going to go away by itself.

On the flip side, as somebody who likes to answer questions, it's hard to find relevant ones. A lot of the more abstract tags don't work for me at all: seeing that a question is tagged doesn't tell me anything about what concrete knowledge is needed to answer it. More generally, I haven't found any good tags to follow, which is the main way I filter through to answerable questions on StackOverflow and Programmers.

As a concrete illustration of the problem, I've written 14 answers and 9 questions as of now, across 49 tags. Not a single one was ever repeated for more than one post, except for the one time when I answered my own question!

I think we need a holistic review of existing tags, with the aim to consolidate and pare them down enough to be more useful. But I don't have any ideas about how to do this that are specific enough for immediate action.

4

What I Disagree

I see where you're coming from, and I'm not agains a big discussion on tags. But I'd like to offer a differing opinion.

  • I subscribe to the , , and tags, and I never feel like I'm not getting what I paid for when those emails come in. The elisp tag suffers a bit of misuse, but that's not a problem with what the tag represents (it's a valid topic, it's just hard to avoid misuse (but I'm diverging, that's another meta topic)).

  • Having a tail of 5-10 question tags doesn't sound terrible for a site barely approaching adulthood. It's the 1-4 tags that we might benefit from reviewing a bit (which is not to say we should kill them all, that's how tags are born).

  • Having tags mean different things is not a problem either. SO has tags for languages, tools, methodologies. The only question we need to ask about tags is: “Can somebody reasonably be expected to be an expert on it?”

    For most tags you mentioned, the answer is yes. Git, org-mode, isearch, syntax-highlighting, are all deep topics where you could expect someone to consider him/herself as an expert. key-bindings is more complicated, the actual notion of a keybind is pretty simple, though the concept of a keymap is actually quite complicated, but this is also a topic of its own.

4

What I Agree

As I said on the other answer, I'm not agains a discussion on tags. In particular:

  • You say tags are difficult to use for answering and asking, I'm very much interested to know if anyone else feels like that.

  • We have a bit of a tag overload when it comes to programming languages. This has been discussed before, and we went with the most logic answer, but I've been questioning myself whether that's the most effective one.


Some of the worries you mention are valid, but they seem more like topics of their own then a single problem.

  • is somewhat useless. It's a perfectly relevant topic, and it even does say something about the question, but it doesn't help narrow the audience. Any Emacs.SE answerer worth his salt knows about define-key, and there's not much more to it. There's a lot of deep knowledge involved in , but that's already a tag of its own.

  • Anyone is invited to have a look at our tail and see if any tags can be merged at all. Just post a topic here on meta if you find any.

  • Are there any specific cases where you found a question difficult to tag?

  • I looked at the tags and it seems that for programming language, the language tag like python is a better choice than the mode tag python-mode in fact there is no python-mode tag. We can start to see inconsistencies when we look at the haskell tags, we have only haskell-mode. It is not intuitive at all. The case of elpy is an interesting one, I guess that elpy should always comes with python tag beside it. Same thing for EDTS and erlang... – syl20bnr Dec 31 '14 at 7:06
  • ... Tags are for experts but also for users and for users discoverability, so the language tag is the most important one and it should be coupled to whatever improves discoverability. Like discovering elpy for python users. – syl20bnr Dec 31 '14 at 7:06
  • @syl20bnr Did you check out the linked discussion? That's sort of what we settled on (always use the language tag, and use the mode tag when appropriate). You're right it's not being used consistently yet. – Malabarba Dec 31 '14 at 17:57
  • 1
    yes I did and this is not the validated answer. The checked answer says that the tag of the mode and the tag of the language are 2 separate things. At least it is what the wording says: choose the apropriate tag. Here I say that the language is what it is the most important, python tag is then a better choice whatever the question is about the mode or/and the language. This is different than the checked answer. For the case of big modes with different naming than the language itself it is good to set both tags. But the language tag should always be set.Same thing for git and magit – syl20bnr Dec 31 '14 at 18:31
1

The way Emacs "solves this" is with apropos and several different apropos flavours. Would it be possible to write an appropos-tags (similar to apropos-documentation) would help people navigate the tag hierarchy, and pick fine-grained tags?

Another related possibility would be to require both general and specific tags, which is how ACM goes about it for the papers they publish.

You must log in to answer this question.

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