What is meta? ×
Meta Stack Exchange is where users like you discuss bugs, features, and support issues that affect the software powering all 128 Stack Exchange communities.

There seems to be a general consensus that we have plenty of tags on SO, and I frequently see some to be cleaned up. A script was recently introduced to eliminate single-use tags after 6 months. More amazingly (to me at least), there's been considerable effort on a tag synonym repository in mere anticipation of a synonym feature.

Would increasing the create-new-tag threshold be another step in the right direction?

How does 4,000 (up from the current 250) to create a new tag sound? It fills part of the void between 3k and 10k, but it feels about right to me. My second choice would be 2k (the same as the edit threshold). However, both are higher than the 500 required to edit tags, which should lead to more users getting experience in tagging before they can create new tags.

For a young or rapidly evolving site, this would not be appropriate, but I think SO has reached critical mass.

Why such a high threshold?

While I don't see any way to figure out who created new tags (besides random snapshots), I do have some numbers:

Threshold  Users
      250  21,561
      500  14,115
    1,000   8,301
    1,500   5,499
    2,000   4,328
    2,500   3,378
    3,000   2,771
    3,500   2,305
    4,000   2,008

Tags: 29,205

While 4k is based on my gut and general experience on SO, I believe going as low as 1k limits effectiveness. In an ideal world, we'd need just one user per topical area and that user would create any new relevant tags, but that ideal user is as mythical as a unicorn. In this world, we only need maybe 50-75 active users per ontology. However, many areas overlap and many of those users would be knowledgeable enough to tag in 5 or more ontologies, since you don't need to be an expert for that.

Once one of those users create a tag, any other user can use it, so this doesn't even require a tag-creator eyeball every page. Furthermore, it doesn't matter how many users the site has: the same number of tag-creators should be able to accommodate them. Much like low-rep behavior is today, users will comment on a tag they feel is appropriate but they can't add to the question. This can be a good thing: in those cases, creating a tag means someone proposed it and at least one tag-creator (at 4k, 2k, or whatever rep) agrees.

Numbers from thin air

If we estimate 40 different areas (probably a high guess for SO) and figure each needs 75 users but 50 of those are shared with four other areas, that gives 40×25 + 40/5×50 or 1,400 users. These users need to be active; so, going by these numbers, 2,008 users (4k) might be too restrictive, but 2,771 users (3k) could be enough.

Related

The limit was increased from 100 to 250 (SO only) almost exactly one year ago.
A recent bug report is [status-declined], but I'm not sure what's being declined. (Does it just mean not-a-bug or [status-bydesign]?)

share|improve this question
2  
@Jon: page=404 made me laugh. (I need another beer.) –  Gnome Jun 2 '10 at 18:09
add comment

10 Answers 10

up vote 10 down vote accepted

Well, let's see.

select t.Name, t.CreationDate, u.Reputation from Tags t
inner join Users u on
u.Id = t.UserId
where Count = 1
order by Reputation desc

For tags with only 1 use:

1,363 total

<= 5000 rep 1086 (79%)
<= 2000 rep  840 (61%)
<= 1500 rep  725 (53%)
<= 1000 rep  529 (38%)

For tags with less than 10 uses:

12,663 total

<= 5000 rep 9745 (76%)
<= 2000 rep 6907 (54%)
<= 1500 rep 5845 (46%)
<= 1000 rep 4363 (34%)

Looks like 1500 rep is definitely better than 1000. I still would rather err on the side of safety here, but if we're preventing ~50% of the bad tags from getting formed, that's a pretty big win while still remaining conservative.

share|improve this answer
    
This doesn't take into account the quality of the tags (no database query could do that). What would be interesting is to get a list of the tag names in the rep ranges (500 <= x <= 1000), (1000 <= x <= 1500), and (1500 <= x <= 2000). I'm assuming the quality of the tags increases proportionally with rep -- it's possible the top 62% of these tags are actually fine. –  Jon Seigel Jun 2 '10 at 19:14
5  
Doesn't take into account the reputation the users had when they created the tags (the experience they've learned in the meantime can be significant). However, that's much harder to query and that they later got that experience may negate some of that. –  Gnome Jun 2 '10 at 19:22
2  
That 1,500 is between two current thresholds is significant, rather than "reusing" either 1k or 2k, as it provides better motivation for users, IMHO. –  Gnome Jun 2 '10 at 19:23
    
I misinterpreted the results and missed that it was tags with a count of 1. My above two comments don't make nearly as much sense, but they still somewhat apply, I think. –  Gnome Jun 2 '10 at 19:27
    
@The Yeah, they're a little confusing. You have to subtract two adjacent values to see how many more new tags would've (in theory) been avoided had the rep limit been that much higher –  Michael Mrozek Jun 2 '10 at 19:28
    
@Jon: Actually, tags with extremely low uses are lower quality: they don't help categorize. Whether counting only single-use tags or any tag with less than 10 uses, I'm not sure which is better. –  Gnome Jun 2 '10 at 19:31
    
@The Cat: I meant that of the single-use tags, those created by higher rep users will be of higher quality. As has been mentioned before, single-use tags aren't necessarily low-quality tags. –  Jon Seigel Jun 2 '10 at 19:45
    
@Jeff: Your updated numbers have convinced me that I'm perhaps aiming too high, but I'm glad you're looking at a target higher than 1k, which I still consider much too low (at least for SO's current user base). The only open thing left to do is make sure tags can't be recreated by lower-rep users when we edit them out (after the site sits for a few days to let the change propagate, etc.). –  Gnome Jun 2 '10 at 21:40
    
Are these numbers after the first run of the deletion script? (Appears so, given page counts.) –  Gnome Jun 3 '10 at 6:40
add comment

From 250 to 4000 seems like quite a leap. Maybe Jeff could modify the script that eliminates single-use tags so that it finds the average reputation of the creators of the tags that are getting deleted? That way we'd have a data point to use to decide where the threshold should be.

share|improve this answer
add comment

I agree that the threshold should be increased, but I think 4k is too high. I also agree with Bill about getting some hard data to support this kind of a change.

I really like the idea of allowing users to be able to retag at 500 and gain experience without being able to create new tags. Therefore, I think increasing the creation threshold from 250 to 1,000 would be reasonable.

share|improve this answer
    
I agree with 1000. Admittedly unscientific, but this feels like the correct number. –  Robert Harvey Jun 1 '10 at 18:21
1  
yes, I think 1000 is the way to go here. I would rather err on making too small of a change in this area, than too large. –  Jeff Atwood Jun 2 '10 at 17:40
    
@Jeff: Could you give us an idea of how much rep people had over the past 3-6 months when they created new tags? (The problem is we kinda need to know which tags to evaluate...) If the problem is people creating too many spurious tags, it'd be nice to know at what point that is reduced. (I don't see how to get that myself, and my random snapshots of the new tag page tell me 1k isn't nearly enough.) –  Gnome Jun 2 '10 at 18:14
add comment

I still look at the 250 rep limit for tag creation as a hold-over from Stack Overflow's early days when there just weren't that many tags; at that point, it was very important to allow most users to create tags in order to seed the database. It's not important anymore, at least on Stack Overflow - it's actually counterproductive due to all of the community effort that needs to be expended on retagging. (Even with a little automation, tag misuse still wastes people's time.)

If we assume that the typical "quality" contributor gets an average of 2 upvotes (or 1 vote + accept) per answer, then a rep minimum of 1000 would mean that you've answered 50 questions, which seems totally reasonable to me for tag creation. Since the rep limit for retagging itself is 500, this also means that any users with tag-creation privileges will have had plenty of time to participate in the "tag management" process; combined with about 1000 rep, that implies some pretty good experience with the tag system.

Of course, this only solves about 25% of the problem. As I've mentioned in the past, the biggest problem isn't somewhat-low-rep users creating new tags, it's the fact that the bad tags remain in the system forever, even after they've been edited out of all questions. Any user at all, even those with 1 rep, can use these tags because it doesn't count as a new tag.

Irritating tags like mssql won't matter when tag synonyms are in place, but who knows when that's going to be, and for the time being, those tags still keep popping up, every day, day after day.

So while I completely agree that the threshold should be increased (and my proposed limit would be 1000, or somewhere around there) - we also need to change the definition of creating a tag to applying any tag that currently has 0 questions, otherwise the problem will continue to linger.

share|improve this answer
    
"It's not important anymore [for most users to create new tags], at least on Stack Overflow" is exactly the way things appear to me. But 4k is still a lot of active users. Even 3k is still every user that has ever cast a close/open vote (and how many times have you felt that number was high?), so 2k is even more. –  Gnome Jun 1 '10 at 17:36
    
[retag-requests] could easily get completed faster than created---with plenty of consideration for each---if tags were created less often, including by plugging the holes you mention. And then with a higher rep threshold, those tags won't keep popping up---but the users will use the tags that exist when not as many can recreate them. –  Gnome Jun 1 '10 at 17:41
    
@The Cat: It's not so much the number of users that's important, at least for something as basic as retagging/creating a tag - it's simply the amount of experience those users can be reasonably expected to have with the tag system. Open/close/delete/undelete is a bit of a different story, you could argue that too many cooks spoil the broth - but I don't think we need a 4k minimum just for creating tags. I don't think the problem is really as bad as people believe - most of the bad tags we see are actually side-effects of the tag-immortality problem. –  Aarobot Jun 1 '10 at 23:34
    
I agree that retagging should have a low barrier, but I think a handful of tag-creators can accommodate an extremely large user base (e.g. a few thousand tag creators for a few million total users). Addressed this more in an update to the question. –  Gnome Jun 2 '10 at 11:16
add comment

This is one of the rare cases where I think a sliding scale makes sense. For example, bumping brand new StackExchange 2.0 sites up to 1500 doesn't make any sense, but it's something I think is long overdue for StackOverflow.

Probably something on a logarithmic scale — perhaps:

required rep = 250 * (1 + floor(log<sub>10</sub>(total tags)))?

share|improve this answer
2  
it would be kind of aggravating for users if the limit was constantly changing, though, wouldn't it? –  Jeff Atwood Jun 2 '10 at 22:36
    
@JeffAtwood - I think this can be solved easily. Instead of having it be a sliding scale, make it a step function. That would make it aggravating, but as sites get bigger, fewer new tags should be needed - there is almost a space of useful tags, and it gets filled, eventually, as the site gets bigger. Once the site is big enough, all tags should exist - the only new ones are for new content areas, and high rep users should be creating those. –  David Manheim Jun 25 '12 at 15:52
add comment

Yes, it should be increased. But 4000 seems a bit too much, and 1000 a bit too less. 2k maybe. But first, I suppose all the single use tags should be deleted. I actually joined meta only so that I could post a question about new tag creation, but found this one. Am seeing too many stupid tags these days.

share|improve this answer
    
Single use tags are deleted after 6 months. –  Arjan Jun 26 '10 at 16:32
add comment

I think that merely deleting single/few use tags is not good enough. Some people create tags arbitrarily that gain popularity. e.g. ubuntu 10.04, and ubuntu 10.04 lts. Actually they should be merged. I wonder if there is something that can be done about it.

share|improve this answer
1  
There's a [status-planned] feature request for tag synonyms. In anticipation of that feature, I created a repository of common tag synonyms so it will be easy to group things together when the feature is released. –  Jon Seigel Jun 29 '10 at 20:13
add comment

I see that the new webapps.stackexchange.com has a limit of only 100 for new tag creation. I think even in the initial stages, it should be at least 200. I already have a 100 reputation on entering there. I'm supposing there is some other reputation exchange program for people who have very high reputation here, so that only they can create tags, and not someone like me. Thanks.

share|improve this answer
2  
there's generally inflation of the 'create new tag' required rep value over time, if history is our guide... –  Jeff Atwood Jul 8 '10 at 3:59
    
@Jeff - I think 150 is appropriate, though, given most any stackoverflow user starts out at 101. Or maybe bump it to 200 after a mere a week or two. –  Joel Coehoorn Jul 8 '10 at 4:19
    
add comment

When my rep was just around 200 I couldn't wait to go beyond that limit and start creating new tags - and then I didn't find any that I need to create.

To me it looks like a lot of new tags are duplicates in the sense of spelling like [x64] vs. [64-bit].

Considering how much I craved for the power to start tagging I now wouldn't even bother with a threshold of 5k.

Mostly it's just a matter of having enough 4k/5k/whatever threshold folks around to re-tag questions with any missing tag.

And as fore those going cissy about that badge ...

share|improve this answer
    
[64-bit], strictly speaking, would include more than just x86-64 stuff -- it would also encompass issues involved with porting software to any 64-bit architecture. –  SamB Oct 20 '10 at 5:28
add comment

If there is a new threshold that would filter out the critical mass of unnecessary tags being created, a query against the Stack Overflow Data Dump would probably find it.

share|improve this answer
    
Perhaps it's just me, but looking at the schema on Data Explorer, I don't see how to determine anything about who created tags. –  Gnome Jun 2 '10 at 11:20
add comment

You must log in to answer this question.