Russell Beattie has a post entitled Spotlight Comments are the Perfect Spot for Tags! where he writes

I read just about every sentence of the Ars Technica overview of OSX Tiger and learned a lot, especially the parts where the author drones on about OSX's support for meta-data in the filesystem. I originally thought the ability to add arbitrary meta data to any file or folder was an interesting capability, albeit not particularly useful in day-to-day activities. But then I was just playing around and saw the Spotlight Comments field that's now included at the very top of a file or folder's Info box and I grokked it! Now that there's actually an easy way to both add and to search for meta-data on files and folders, then there's actually a reason to put it in! But not just any meta-data... What's the newest and coolest type of meta-data out there? Yep, tags! And the comments fields is perfect for this!

Obviously nothing has changed in terms of the UI or search functionality, just the way I think about meta data. Before I may have ignored an arbitray field like "comments" even if I could search on it (haven't I been able to do something similar in Windows?). But now that I "get" tagging, I know that this isn't the place for long-winded description of the file or folder, just keywords that I can use to refer to it later. Or if those files are shared on the network, others can use these tags to find the files as well. Fantastic!

This sounds like a classic example of "When you have a hammer, everything looks like a nail". One of the interesting things about the rush to embrace tagging by many folks is the refusal to look at the success of tagging in context. Specifically, how did successful systems like get around the Metacrap problem which plague all attempts to create metadata systems? I see two aspects of the way applied tagging which I believe were key to it becoming a successful system.

  1. Tagging is the only organizational mechanism: In, the only way to organize your data is to apply tags to it. This basically forces users to tag their data if they want to use the service otherwise it quickly becomes difficult to find anything.

  2. It's about the folksonomy: What really distinguishes services like from various bookmarks sites that have existed since I was in college is that users can browse each other's data based on their metadata. The fact that is about sharing data encourages users to bookmark sites more than they typically do and to apply tags to the data so that others may find the links.

Neither of the above applies when applying tags to files on your hard drive. My personal opinion is that applying tagging to the file system is applying an idea that works in one context in another without understanding why it worked in the first place.


Tuesday, May 3, 2005 5:56:27 PM (GMT Daylight Time, UTC+01:00)
The problem with tags used in this way is that you can't tell when a text string is actually a tag. On a small filesystem this isn't a big deal but on the Internet it's obviously a problem.

There should be an explicit field for tags.
Tuesday, May 3, 2005 7:57:32 PM (GMT Daylight Time, UTC+01:00)
I find this intensely amusing as OS/2 had the capability of adding arbitrary "attributes" to files back in 1990.
Wednesday, May 4, 2005 1:02:23 AM (GMT Daylight Time, UTC+01:00)

"...without understanding why it worked in the first place."


I'm quite aware of how and why tagging works, thanks, though I'm not sure you do. Regardless, I was simply exploring the idea of local tagging and thinking out loud about how it would apply to my local and/or shared files. Why you're naysaying even contemplating the idea is beyond me.

This is like the trifecta of obnoxious weblog posts - Not only were you arrogant and condescending for no reason other than annoy me, I hate the word "folksonomy" as well. I'll look around for a prize for you.

Wednesday, May 4, 2005 4:39:53 AM (GMT Daylight Time, UTC+01:00)
I agree -- this is fundamentally different. The driving factor behind the popularity of the current social tagging services is that other people do most of the tagging for you. You might tag 100 things, but you get the benefit of (100*users) tags.

On your own filesystem, you get only the benefit of the tags you created. This doesn't scale, and quickly becomes an organizational burden. Think of all the other "self-tagging" solutions that have bombed: tags in your photo browser of choice (for your own digital pictures,) descriptive tags on your MP3s (not including those that were automatically generated,) and those damn DESCRIPT.ION files from the BBS days.
Wednesday, May 4, 2005 11:25:34 AM (GMT Daylight Time, UTC+01:00)

I have a question for you:

I get the feeling you are assuming that Tagging is something extra that happens? If you think like this it is almost self evident that people will avoiding tagging because it is extra work.

But isn't putting a document in a folder path called ...\Personal\Mum logically the same as tagging it with both Personal and Mum?

If the file system actually treated folders as tags, then tagging would be no extra work. Since you do it when you choose where to store the file.

Why is this useful?

You would be able to find that document in all the following places

I.e. you only have to remember one of the two tags to find it. Compare that to the standard file system. You not only have to remember both Mum and Personal you have to remember their order of precedence. There is a lot of research that indicates that people find things by starting somewhere they know and then narrowing the search, and this type of filesystem would definitely supports this.

I'd also like to draw your attention to the fact that if your folders are Tags... then at least one of your criteria for success is satisfied:
1) Tags are the only organizational metaphor.

And looking at your second criteria, why do you social need encouragement to tag? If folders are tags you are doing no more work anyway?
Thursday, May 5, 2005 5:35:48 AM (GMT Daylight Time, UTC+01:00)
Alex James nailed it for me. That scores of other people are also tagging bookmarks is nice, but this an aspect of I rarely use. I like tagging because it makes it snake-simple to find things later on.

If I bookmark a page about, say, JavaScript+CSS, I don't have to fret over what boomark folder it goes to. It goes "into" both. Or more.

The use of tags also helps me see related items.

These are exactly the types of things I want for my local file system.

Thursday, May 5, 2005 9:45:47 AM (GMT Daylight Time, UTC+01:00)
Tags in the filesystem are not strictly necessary, but they're only half the story in Spotlight.

The order-of-precedence argument is bogus, as there is nothing to stop you having a directory called `mum_personal' instead. Or alternatively, you could have a designated index-file in the directory.

IMO where Spotlight wins is the ability to search exif data in images, and suchlike obscure formats.

(Note that there's nothing that said such formats could not have been designed with multiple files for separate meta-data, so your good ol' find and grep commands would've worked just fine.)
Friday, May 6, 2005 7:26:41 AM (GMT Daylight Time, UTC+01:00)
>The order of precedence argument is bogus, as there is nothing to stop you having a directory called 'mum_personal' instead.

Yeah why not flatten all you subdirectories into long one directory name, sounds like the answer to every problem ??
Friday, May 6, 2005 10:55:06 AM (GMT Daylight Time, UTC+01:00)
> Yeah why not flatten all you subdirectories into long one directory name, sounds like the answer to every problem ??

I hope you realise the irony of this, in that this is exactly what tagging does - supply a flat identification system. Next thing you know, someone will decide they *want* hierarchy on top of their flat world. But the filesystem gives you the options for both. Either way it sounds more like a feature of the search system used, than the backend db. (There's nothing to stop you writing a graphical variant of find(1) that treats directories as tags irrespective of order, if you want!)
Saturday, May 7, 2005 12:13:45 PM (GMT Daylight Time, UTC+01:00)
Tim, I hear what your saying and now you point it out I do see the irony! Never was the quickest of cats.

I suppose your point on it being a feature of the search system is correct... I think you are hinting at this -> when a file is has a path like named X\Y\Z\file then it is tagged with X, Y and Z so when searching I can quickly retrieve it by using just the X tag instead of the whole path.
We should probably carry on this little discussion via email since I'm not sure that Dare put up his blog comments for us carry on a conversation.
Monday, May 9, 2005 7:18:56 PM (GMT Daylight Time, UTC+01:00)
So WinFS is a waste of time then?
Thursday, April 20, 2006 4:13:08 AM (GMT Daylight Time, UTC+01:00)
It takes all kinds of people to run a world, and if we didn't have people like Tim then we wouldn't have any sheep to shorn nor lackwits to enslave.

First, every system is exactly like every other system if you squint hard enough. Thus, filesystems are "exactly like" databases if only blah blah blah. Of course, this completely misses the point that every single interesting thing in computer science is a consequence of human psychology. And human psychology puts a brick right through the idiot notion of "squinting hard enough". (Feel free to ignore this advice Tim, as a programmer you're expected to ignore practical reality.)

Second, the order of precedence argument is far from wrong. When you actually have to categorize a few hundred to several thousand novels, sets of photos (yes, hundreds of *sets* of photos) and movies, the combinatorial explosion becomes impossible to deal with. Does this movie go under lesbian or whips or black? Does this novel go under highlander, romance or extra good?

Third, it is in fact ALL in the searching. And it's obvious that neither Tim nor Alex James understand the context in which delicious works. Well here it is, so listen up and pay attention folks.

Delicious works because searching is AUTOMATIC and BRAIN-DEAD. And I don't mean searching for objects that have already been tagged. No, what's automatic and brain-dead is searching for tags to categorize stuff. And a little wee thought into the problem ought to enlighten everyone about why this matters.

It should be obvious, at least in retrospect, that anyone that amasses a large database of objects stores 10x more objects than they'll ever retrieve. You retrieve 10x more objects than you'll ever store, but the objects you retrieve are invariably OTHER people's objects and so they don't count.

The average user reads, watches, or listens to 100 objects that belong to OTHER PEOPLE, categorizes 10 of them, and then later retrieves just 1. This is why "searching" for objects simply doesn't matter. What does matter is automated searching for tags. This is what delicious is all about.

So until you have a filesystem that automatically makes suggestions for tags based on tags that already exist, object type, filename, and any other remotely relevant information, you have nothing even remotely like delicious, only something that is "exactly like" delicious "if you squint hard enough".
Comments are closed.