15

I wonder how to choose good tags on a new Stack Exchange site.

On the one hand, this is very much a community thing: askers should tag their questions with the tags they'd expect to see if they were searching for the question.

On the other hand, tags classify, which is hard to do if you don't have a good idea of the whole picture. So it's up to more experienced people to pick good tags. Also, having a small group pick tags helps reduce the number of near-synonyms that can't be sorted out automatically.

On the gripping hand, that sometimes leads to tag names that many askers or searchers simply don't understand or think of.

I wonder what people's experience is on the various SE 2.0 betas (or from the early days of the trilogy sites). How do you come up with good tags (beyond the obvious like programming language names on SO)? Do you create very specific tags early on wait until later to split large, imprecise tags? Are tags mostly chosen by a small experienced group (so you get better consistency) or by the wider community (so tags may reflect people's expectations better)? Any tips for burgeoning SE 2.0 sites?

I've done a lot of tagging on Unix, some on SF&F, and I still don't really have an answer to offer. Any advice? please?

3
  • Is this a question about how to choose tags for a new site that you're creating, or about how to choose tags for a new StackExchange site? If the former, I think this is more for Programmers (or SO). If the latter, then I'm not sure how to answer... Apr 30, 2011 at 23:40
  • @DavidThomas: It's about SE sites, which is why I'm asking here. Originally I had Unix in mind, now it's SFF, but I'm after general advice and experience reports from other SE2.0 sites. Apr 30, 2011 at 23:55
  • Ah, then it makes sense that it's here on Meta... =) (I still have no idea how to answer it usefully, but I do appreciate the response...) Apr 30, 2011 at 23:56

4 Answers 4

10

My guidelines are simple.

  1. No meta-tags.

  2. Each tag must potentially work as the only tag on a question, if necessary. If it cannot, it is a bad tag.

  3. Each tag must have a clear, unambiguous definition within the context of that community, plus or minus 10 percent. If a reasonable group of community members can't agree on what the tag means, it is a bad tag.

  4. Tags should be a meaningful categorization of content. If you cannot imagine another human being ever wanting to browse in or subscribe to that tag ... it is a bad tag.

  5. We do not support tag hierarchies. If you are attempting to ghetto-hack hierarchy into our engine with your tag, it is a bad tag.

  6. Tags are for categorization of broad topics, not minutiae. If your tag is duplicating the work someone could do with a trivial search or tag + search, it is a bad tag.

1
  • Nice set of rules. If only it was clear how they work in practice… Re 3: what about tags that are meaningful to experts, but where more than 10% of askers don't know the difference? Re 4: so a tag that is useful to search in but not to browse in or subscribe to is bad? Re 5: I have no idea what “ghetto-hack hierarchy” means. May 5, 2011 at 20:33
3
+200

I've been doing a lot of retagging on UX lately. I don't claim to have any expertise* on the matter, but here are some thoughts:

  • Coming up with good tags isn't hard, since lots of people participate in that process. The majority of tags are introduced by someone asking a question.

  • The most popular tags will tend to be "meta tags" that are too general to be useful. Lately, we've been eradicating the meta tags from our site, which has left us with a bunch of untagged questions.

  • We probably waited a bit too long to start killing meta tags. Retagging hasn't been hard, since we now have a long list of useful tags from which to choose. I think you should identify and ban meta tags once replacing them with existing (non-meta) tags becomes easy.

  • I rarely create a new tag myself. I won't do it unless I can match it with at least three questions.

* Although along with English, UX is one of the best places to find true experts.

4
  • That's interesting, because this doesn't blend well with my experience on U&L and SFF. Maybe I should try to write it up after all… For example, SFF has tended to converge towards author names and a few meta-tags, which is not helpful. Using tags for each technology (e.g. a program name) is more useful on U&L, even for technologies with a single question (it'll grow eventually). Do you think the UX community may be better at choosing tags? May 2, 2011 at 18:16
  • 1
    @Gilles Why aren't author names helpful? May 2, 2011 at 19:20
  • Apart from the most popular authors, people won't ignore or favorite these tags due to the low volume, and they're not useful for searches because the name would also work in a plain text search. I'm not opposed to using these tags, but I think additional tags for subgenres or themes would be useful as well. (Yeah, I should really write that answer. Give me time to go through non-SE and non-Meta stuff first.) May 2, 2011 at 19:48
  • There, I've posted my answer (where it's pretty clear that I don't really know where I'm going). You're the only one in line for the bounty so far, but I'd really like it if you could expand your answer. In particular, do you think your UX experience is generalizable? May 4, 2011 at 22:45
2

So I'm going to ramble about my experience with tags on two sites so far. TL,DR: yeah, probably.

First, I see two and a half uses for tags:

  • People who're interested in answering (or more generally seeing questions) about certain topics can ignore or favorite certain tags.
  • People who're searching for something can narrow their search.
  • Tag wikis (but do people actually read them)?

Unix & Linux

Most tags are names of applications or technologies. They're what people naturally use when they want help on a particular technology, which is what most questions are about. A question about using KDE? tag it kde. Some of these tags have lots of questions (for example linux is the biggest tag by far), others have very few (sco isn't exactly popular), but it's pretty clear that these are desirable tags. There's the occasional name coincidence hurdle (e.g. is ports about TCP or BSD?) but nothing we can't easily solve on a case by case basis.

Then there are topic tags, such as security or networking. Those have mostly arisen naturally as well. They're what people tend to use when they want to do something but don't know what the right tool is. I'm still sometimes undecided as to how precise to go; for example networking is pretty wide, and I'm still undecided as to whether it should be used when more specific tags apply (firewall, ethernet, …).

Sometimes the distinction between tags is subtle. For example, what's the difference between shell, command-line and terminal? Beginners will often use shell and terminal indiscriminately, but at the level of U&L, we do want to make the difference (shell is our second-biggest tag, and bash (a kind of shell) is close behind). Since the difference useful but subtle, we don't expect askers to always get it right. In general, though, I don't know where to stop caring about differences that are only apparent to experts (should x11 be separate from xorg? packaging from package-management?).

Hyphenated tags — tags about a frequent combination of two distinct concepts — arise rarely, but in high-profile topics. Should Linux kernel questions be tagged linux-kernel or linux kernel? (we settled on two tags). Should shell scripting questions have their own shell-script tag, or do we combine shell and scripting (you can shell without a script and script without a shell, so they're not synonyms in any way)? we haven't been able to make up our mind.

Meta-tags haven't been a problem. There are borderline cases, like software-rec, which characterizes the nature of a question rather than its topic (but I find it useful: it marks the nature of a question succinctly in search results or in a feed). And each site seems doomed to have its own debate about homework. But no major issue there.

All in all, tagging for Unix seems to have worked reasonably well, with just a few hurdles.

Science Fiction & Fantasy

Most questions are about a specific work, author or universe. And most of these questions have just one tag: the name of the work/author/universe. Sometimes with variants such as lotr j-r-r-tolkien. We're still undecided as to whether that's generally useful: after all, the name should be in the body and so the question will turn up in searches anyway. For popular works, and especially universes spanning multiple works, these tags are useful (a star-wars question might only contain the name of one of the movies in the series, and that's a tag fans subscribe to). But at the other end of the spectrum, who's going to subscribe to <obscure author name>?

These names tend to lead to near-synonyms that aren't actually synonyms. Discworld is Terry Pratchett's best-known work, so I expect most questions about him will be tagged discworld. Does that mean terry-pratchett should be used on all Discworld questions? or only on non-Discworld questions? We haven't yet had enough volume in a tag to need to decide.

We've just started to encourage topic tags. If your question is about spaceships in Star Trek, tag it star-trek spaceship. I hope it'll catch on. That would leave only questions about specific plot points or characters tagged solely with a work or author name.

It's been suggested to use subgenres as tag names — similar to English Language & Usage's broad tags (word-choice, verbs, …), I suppose. We practically haven't been doing that. I think the reason is that subgenre classification is awfully vague and subjective. Can tags still work if people don't agree on what they cover? Are subgenre names meta-tags for us?

Or is the proper analogy with EL&U to use tags like character, plot and setting? While these are objective, I don't see what value they bring to our site. Is it really useful to quickly sort between Harry Potter character questions and Harry Potter setting questions?

We have a couple of meta tags. The ones that remain are list and recommendation, both marking types of questions that have been rules off-topic and are all closed (and that we might delete en masse at some point).

I don't feel we've settled on good tagging practices for SF&F.

Conclusion

I find it interesting that Patrick McElhaney's experience on UX is pretty different from mine, and mine is different on U&L and on SF&F. My impression so far is that different communities face different problems when choosing tags. So, can you, gentle reader, find common patterns and come forth with good advice?

1

I've been thinking about this recently and wondering if perhaps the role Area 51 has to play in the "definition" phase might be able to help with this.

Perhaps when questions are proposed as a "great on-topic example", and "great off-topic example" Area 51 could request tag suggestions for said question? This would then feed into the site if and when it reaches the Private Beta. Good tags could be pre-populated, to help build-out the tag set and provide a starting point for people to work from. Bad tags could be "soft-blocked" so if a question is tagged with one during the private beta, the system flags up that this was considered to be a bad tag and therefore recommends against using it.

Area 51 provides a great way to try and define the initial scope/feel of the site, and this would be a fairly simple (says the person who wouldn't have to code the changes! =) way of helping guide a sites taxonomy from an earlier stage.

You must log in to answer this question.

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