0xDECAFBAD - Tag: flickr

Tags do work (for me, at least)

2009-01-18T06:29:34+00:00

I've been using a lot of tags on Delicious over a relatively long time, so they seem very useful to me.
Delicious encourages the use of tags through UI convention and tool usage patterns, whereas Flickr presents no particular bias toward collecting tags from users.
Since title and description attract more contribution effort from users on Flickr than on Delicious, it's natural that search over those fields will be more productive than for tags.
Search on Delicious doesn't have access to the complete text of the bookmarked resource, and often tags will contain information missing from the supplied title or description.
All told, tags on Delicious are more essential than tags on Flickr.
In conclusion, I think Do Tags Work? misses the value of tags, as I know them, by focusing on Flickr.

Of course, I don't really care what this means for folksonomy and the rest of Web 2.0—tags work for me on Delicious. So, I suspect this means I'm not entirely opposed to the sentiment in Do Tags Work?, because I don't think tags work everywhere their use is attempted.

The rest of this entry elaborates on the above.

From Cathy Marshall's Do Tags Work?:

Have I convinced you that tags aren't all they've cracked up to be? I hope I have, but nonetheless there's a lingering fascination. Surely there's something to be done about tags: we don't want to just turn up our noses at Mr. Weinberger's argument.

From Mark Bernstein's In TEKKA: Tags:

The study could be repeated on more images, and in additional contexts. It might be bad luck. But if it’s a repeatable result — and I think we all know it is — then we’re going to have to rethink a lot of our Web 2.0 rhetoric. Folksonomy is an illusion.

From my bookmark on Delicious:

If you can get past the rambling paragraphs of awkward fun-poking at tags interspersed with library science / web 2.0 / cultural references—as well as a discovery of what, you know, Flickr is all about—there's a well-embellished and obsessively-assembled statistical analysis of tags vs title vs notes in finding photos featuring tourist heel-spinning on the testicles of a bull mosaic in Milan. My impression is that she's missed the point of tags, but I'm having trouble reducing the impression to a critique.

It might be that I'm intimidated by the writing in Do Tags Work? or by the fact that a lot of work was done to produce numbers I can't dispute. Either way, I have neither the time to produce a counter-study nor the wherewithal to extract and refute points in that article with precision. So, I'm just going to throw my impressions out there and hope they coalesce.

Tags are extremely useful to me.
- I've been using tags on Delicious since shortly after the site was created.
- To date, I've saved 11825 bookmarks and spawned 6069 tags. (More on that ratio later.)
- Tags have worked very well for me, both in later finding my own saved resources and in finding aggregates of others'.
Flickr is not the poster child of tagging—Delicious is.
- On Flickr, I've found tags occasionally interesting, but not incredibly so. More a curiosity than a killer feature, really. (More on that in a bit, too.)
- To be honest, I've not found a great deal of use for tags around Web 2.0 properties beyond Delicious, folksonomy hype notwithstanding.
- Fun fact: When I worked on Delicious, we disliked the term "folksonomy" and tried never to use it in serious discussion.
The places where I've found tags useful are text-heavy, where the full text is either not indexed or not displayed in results when using search.
- Consider magazine articles or blog posts bookmarked on Delicious.
- Those bookmarks are subject to the character limits of title or description fields.
- Far more often than on Flickr, there is information in tags not present in a bookmark's title or description.
On Delicious, the use of bookmarklets and extensions is the primary source of new bookmarks.
- These tools tend to produce a title based on the original bookmarked page and, when present, a description summarized from the page contents by way of highlighted text.
- With this in mind, consider that tags often make up the sole intentional contribution made to the bookmark by the person saving it.
  - Beyond the choice to save the bookmark in the first place, that is.
- Whereas on Flickr, the effort is spent on title and, sometimes, the description. Beyond that point, tags are an afterthought, at best.
The user interface of tagging is near effortless versus, say, formal taxonomies or nested folders.
- Even the much-debated choice of the big, fat space bar as a delimiter supports minimal effort.
- Minimal effort lowers mental cost, which promotes more noise, but also encourages more input (or any input) and thus more signal overall.
- People have limited attention span budgets from which useful metadata can be solicited. Tagging lowers the budget requirement.
The reason why I have generated a ratio of 1 new tag for every 2 bookmarks is that I use tags in a way akin to free association, made possible by the low cost of tagging.
- This is a feature, not a bug.
- I have many, many tags used only once and never again—some of them are, in fact, jokes.
- But, I have generated some tags that I've come to adopt more heavily, most of which proved surprisingly useful over months or years of recurrence.
- As for those tags I've never used again: They cost nothing, can be hidden below visibility thresholds, and can be merged later into more appropriate tags if I care to do so.

And, there's where I run out of steam. For the conclusion, go back to the beginning of this entry. Biased though I might be, I think Delicious is the place to study tagging, and there they're of great use. This is where I think Do Tags Work? misses the value, despite the volume of writing and data.

Archived Comments

kellan
2009-01-18T07:40:08

In addition Flickr's tagging implementation has always been non-optimal because there is a 1-to-1-to-1 relationship of object (photo) - tag - tagger. To call tagging on Flickr a folksonomy is to fundamental miss why the term was coined to describe del.icio.us (we can argue about whether we like the term). Tags in del.icio.us act as votes (among other things), much like links in PageRank, and the mish-mash of curatorial activity on Flickr (faving/commenting/groups).

Which is a long way of saying I agree! (and we're thinking about how to fix tags on Flickr, but thats been true as long as I've been there)
Mark Bernstein
2009-01-18T14:17:50

It seems to me that it ought to be straightforward to repeat this study of delicious.

If I follow you, you find delicious tags useful primarily as a personal mnemonic -- a way for finding your own stuff. That's an valuable application, but not the one addressed here (or by folksonomists); Marshall is exploring whether tags are useful for helping the community organize a collection for collaborative use. I haven't heard much about using delicious this way, but perhaps it could be done.
karl
2009-01-18T18:30:05

note that she is saying that tags are good at finding your own personal stuff, and that you confirm this in your blog post. So at least you are in agreement on this.

But let's dig a bit further what is she saying and what you are saying. I have too tagged a lot of stuff with another service similar to del.icio.us. The result is that no, it is not useful to find something quickly, and I would even go further that I don't even find my own stuff. What I do though on a regular basis is to reduce my own stuff by changing the tags for a set a of pages for example if there are économie, economie, economy, business, I might finally reduce it to one tag "économie". What I would really wish is that I could do a hierarchy of these. For example sometimes I tag Japan and Tokyo, what I wish is that in my own hierarchy when I tag Tokyo, the system assumed Japan.

On Flickr it really depends on what we are looking for, but I'm using flickr for finding images for specific location and for specific creative commons licenses. In both cases, these metadata depends on an ontology (taxonomy, structured schemas, $ORGANIZATIONSYSTEM)

geolocation… the ontology is the map, a well defined system where you can pinpoint an image somewhere and get the real lat/long values.

licenses… a controlled vocabulary of different type of licenses.

The debate is not about "do tags suck" but do they work? And indeed they are not really efficient compared to other methods. It's cool they exist, but they should not be here to justify that ontology sucks either. Every community which is in a need of achieving a specific repetitive task comes one day or the other to structured data organization.
Ethan
2009-01-19T02:33:33

Very often, the suggested tags on a new Delicious post inspire my personal informational ontology. Folksonomy is real and valuable, but like any public resource, it requires some effort on my end to make it worth my while. Flickr's tagging system is valuable to me, but not as crucial to the overall experience. This is in part because Flickr's thumbnail images are a faster way to free-associatively group like concepts together than verbal tags ever could be.

Your readers might enjoy an analysis of how I use Delicious.
l.m.orchard
2009-01-19T02:37:53

@Kellan: That's definitely another thing - Flickr has plenty of explicit curatorial features that suck up the energy otherwise directed toward tagging—the only game in town on Delicious.

That's not a bad thing - those are good features that wouldn't necessarily work for Delicious. And if they would work for Delicious, someone somewhere is probably emulating them with specially-formed tags to varying degrees of success. Eventually we might have discovered those uses and wrapped a feature around them. (see also: for:* and system:media:*)

Actually, one area where I think Flickr has done better with tagging is in machine tags. There was some disagreement about that even on the Delicious team—pitting it against simplicity of tags—but I thought there was something there worth exploring.
l.m.orchard
2009-01-19T02:41:29

@Mark: Not primarily as a personal mnemonic—I probably should have expanded more on the social aspects. I find tags on Delicious just as useful for finding others' bookmarks, too.

I assume that many others use tags much like I do—especially the people bookmarking things I find interesting. On Delicious, something akin to lightweight communities emerge based around bottom-up tags.

I am no library scientist—but, if you've got a community with a collection to organize, I don't think you really need tags. That community probably already has a notion of categories worth extracting in a more formal top-down process.

As for repeating the study, I probably won't be the person to do it. But, for a quick example that I'd suggest is par for the course, take a look at the bookmarks for the article in question.

Of the 18 bookmarks, 10 have notes. That page only shows the most common title, but I'd be surprised if any of the 18 have a title other than "Do Tags Work?" That leaves the tags. Of the top 10 tags, only one appears in the title ("tags"). The word "folksonomy" appears in none of the descriptions. There's also "socialsoftware" used 3 times, which appears neither in title or description, nor in the original article itself. I think this shows better use of tags than the example used from Flickr.
l.m.orchard
2009-01-19T03:12:10

@karl: I wrote that tags are useful for finding both mine and others' bookmarks. Meaning tags work as intended, at least for me on Delicious. I suspect they work for others too, since they keep tagging things in ways that I keep finding them.

You wrote about another service, but I didn't—I'm writing about tags on Delicious. I don't know what the thing they call tags looks like on that other unnamed service.

This is part of the reason behind my mild disdain for the term "folksonomy": Delicious didn't introduce tagging to start a best practice for Web 2.0 at large—tagging was introduced as a site feature for Delicious. Attempts to clone that feature elsewhere have been uneven in results, despite having had a term coined and a body of hype spun around it.

Tagging is like a salt water fish that lots of people thought was pretty and they started trying to stick in fresh water tanks. I don't think it thrives everywhere people have tried to stick it and not everyone who's tried to clone tagging has gotten all the important parts right.

On Flickr, I've asserted that tags aren't that useful. Machine tags like geo tags are a big exception though. Otherwise, they have a wealth of other tools for finding things beyond tags, whereas on Delicious they're the only game in town.

I guess my main point is this: In my experience, tags work. On Delicious. Where they're largely agreed to have been cloned from. Elsewhere, I think there have been transcription errors and misinterpretations that give the concept a dubious reputation.

I'm not sure what you mean about tags being not really efficient compared to other methods. Tags are a way to get people to provide metadata useful in emergent classification that would otherwise have not been supplied at all. I think they're efficient at that.

Many people just punt when faced with sorting a new resource into a formal category or folder structure. I suppose that filters out all but the conscientious, leaving better quality results. Is that what you mean?
Daniel Tunkelang
2009-01-19T06:07:24

My colleagues and I at Endeca learned a lot from out experience with the ACM Digital Library, a collection of author-tagged computer science articles. As is, the author-supplied tags weren't all that useful. But we were able to derive a vocabulary from them that we then used for automatic tagging. We also applied a variant of this approach to the website for a leading sports programming network.

You can learn more details about our approach in the proceedings of the 2008 Workshop on Human Computer Information Retrieval:

http://research.microsoft.com/en-us/um/people/ryenw/hcir2008/
Britta
2009-01-19T06:35:48

Heh, so much of this depends on personal use. Like we always tell people, there's no wrong way to tag, and that makes it really really hard to generalize accurately about What Tags Mean.

I tend to use Delicious tags to associate a bookmark with a thread of my thinking (which is very useful), and I use a combination of search and tags to find things. But on Flickr, I like looking through tagged photos, and I like looking through groups even more. I search to find old photos, often finding a picture because of the tags on the photo. Both Flickr and Delicious work very well for me.

Tags are interface; the "meaning" of each tag is de-emphasized. Compared to hierarchical structures, they're easier, messier, more-used, more fun. They're flexible - they're personal.

Tags on Delicious are collective; tags on Flickr are collective (except when you tag another person's photos); folksonomy implies collaborative.
Britta
2009-01-19T06:39:57

P.S. I rewrote pretty much the entire Wikipedia article on tags a few months ago - http://en.wikipedia.org/wiki/Tag_(metadata) - which was fun. If you disagree with me about tagging, just look at the official article! Ho ho ho.
Paul Mison
2009-01-19T13:17:34

Just to throw a minor spanner in Les's argument in comment #5, I do have a slightly different title: I put the site name after a pipe for every bookmark I make, partly because it might make it easier to distinguish posts from different sources at some point (without recourse to URL or tag) and partly because it looked good when Dan Hill did it (I nicked the idea, as with so many things.) I don't think this really invalidates his point, though: tags are the place for (most) people to put their mark on a link. Us description-editors are odd.

(Annoyingly, I can't find a link that shows that well. There's my bookmarks tagged 'tagging', I suppose. I've only just noticed this has related tags- that's a nice touch I'd never really spotted.)

On a side note, I love machine tags in Flickr, mainly because they're hidden from the user by default. (They've got a lot more useful now that there are specialised methods in the API, too.) I mused on Twitter last week that neither Delicious nor a third party has done anything with the loose convention of via: or cite: tags that some people use. One day...
Karl
2009-01-19T17:52:28
I said tags were not very useful to me in the sense they don't help me to find interesting stuff (even on delicious) :) Your mileage may vary. Understood.

About categories and structured hierarchies, I never mentionned to have to pick up one unique category. My goal is to be able to have hierarchieS. You could imagine a system where one tags "chinese"
1. meant a human chinese person
2. or meant the written language.
Structures are useful. As I said, I'm not against tags, I'm just for a better use of tags. I believe that plugging SKOS would be a very easy way to leverage the power of tags by letting people doing their own personal ontologies if they need it. :)

So the summary instead of a flat system, I want a hierarchical system, a graph without destroying the flat system.
l.m.orchard
2009-01-19T18:13:37

@karl: I'd say once you introduce intentional structure or hierarchy, you no longer have tags. I think you really want categories. You can emulate categories using tags, but you'll be going against the grain.

Tags are about throwing things at the wall and letting analysis and algorithms help surface implicit structures or relations.

And, yes, the semantic issue of "chinese" the language vs "chinese" the food vs "chinese" the people is a long standing one. Single tags don't solve this and tagging in general kind of shrugs on the issue—though tag intersections involving many tags help disambiguate intended meaning. (ie. chinese+food, chinese+language)
My tag cloud
2009-01-19T20:31:25

I didn't thoroughly read this post or the link, but I'm going to guess that the value of tagging depends on the tags you use. For instance, I wouldn't tag this with "tags" but instead with "tagging debate", "tagging meta", or similar. That would allow a visitor to this site to find every similar discussion. The "tags" tag might include a lot of things that someone wasn't looking for, such as a comparison of tagging services.

This could also be tagged with the name of the person who wrote the link so the visitors could find discussions involving that person.

For how I do things, my tag cloud is at my name's link.

Note the tags with [name topic], such as "obama cabinet". Anyone who sees that should realize right away what it involves. And, someone looking for specific Obama information can scan the tags starting with his name for subtopics. Note also the tags about themes, such as "mean spirited", which refers to an "argument" some people make. And, click a tag like "bill richardson" to see how I make some sense of large numbers of posts: many of the lists with a large number of posts are preceded by a summary linking to individual posts.

(Note: that doesn't represent all the content at my site because I switched from MT to Drupal a couple months ago and only a fraction of thousands of posts is tagged. There are also a few hundred tags that aren't shown; I need to figure out how to show lots of tags.)

Don't ask me who I am

2006-08-17T17:24:25+00:00

I'm tired of filling out profiles, and I'm tired of building up networks of others' profiles online. Everyone needs to adopt something like ~~FOAF~~ hCard and XFN (updated per suggestion from Kevin Marks), and be done with it. The networks should be semi-automated, or at least suggested, by detecting who I know from other places as they hop on another bandwagon with me. (The separations between implicit groups bounded by the particular service are as important as explicitly defined groups.)

LiveJournal was my first experience with much self-disclosure online, into which I dove with much enthusiasm. But since then, I just haven't felt the need to replicate self-disclosure anywhere else but on my own site. Orkut, Rize, LinkedIn, MySpace, Vox, del.icio.us, last.fm, flickr, 43 {Things|People|Places} -- these are sites in which I've had some passing or lasting interest in the last few years, but I wish I could just point them all at a single source for learning about me. It all needs to be turned inside out, and lock-ins removed.

Centrallized identity and social network management for distributed service providers is what I want. I'm sure there's already a movement afoot, and I've just not reached that spot in my to-read list.

Archived Comments

Mad William Flint
2006-08-17T18:42:12

bah. they should take vCard data and be done with it. So much already does.
l.m.orchard
2006-08-17T19:47:00

Well, one thing I like about FOAF, is that it's extensible. So, as more categories of self-disclosure become popular, they can be dumped into that one central thingy
Geof F. Morris
2006-08-17T20:23:01

Actually, Les, I haven't seen it either. I'm with you: I sure as hell would love to only write this crap once. Even copy-and-paste is just too much shit to do!
Bob Zoller
2006-08-18T00:38:27

This is basically Marc Canter's current mission in life, from the sounds of it. See his PeopleAggregator (http://peopleaggregator.com/) project.
Mad William Flint
2006-08-18T01:43:17

Foaf's cute. but I don't think it's any more extensible than vCard is (iirc)

However, it is nice for actual F-O-A-F semantics. But you'll never see someone's foaf at the bottom of their email, auto-imported into exchange, etc.

Maybe I'm a dinosaur, but there's all ready too damn much xml in the world by my reckonin'
Martin Spernau
2006-08-18T07:49:15

Yes, the Peopleaggregator sets out to solve this :) This http://traumwind.de/tindertraum/archives/debabelizeing_profiles.html is the beginnig of the "IdentityHub" -Martin
Terrell Russell
2006-08-18T13:45:20

claimID, claimID!

http://claimID.com

Give it a whirl and see if you like what we're doing. Point to it. Point from it. Keep it in one place.

Terrell http://claimID.com/terrell
Mad William Flint
2006-08-18T14:51:47

Nice. I'll concede the point to hCard.
Jim Renaud
2006-08-18T16:48:39

Can't we just use the information the NSA is compiling for us? This way we don't even have to enter it ourselves!
Ian Bicking
2006-08-18T16:51:10

http://videntity.org/ seems to be offering a bunch of these things together; more as an example, since anyone offering the same information would be a peer.
Bob Aman
2006-08-18T23:34:44

I'm a minor fan of claimID, but I much prefer hardcore decentralization. FOAF would be a lot more useful if it didn't rely on RDF. XFN and hCard are good for what they are, but they're still a far cry from the expressivity of FOAF. Sadly, there's no good format that solves a sufficiently large number of the problems in this space (that doesn't carry the RDF baggage anyways).
Danny
2006-08-19T09:01:02

Right on Les! There are two issues, identification/authentication (hard, but the outlook looks promising) and personal profile/description (easier, already pretty much solved by the tech you mention). Hopefully the two should be joined up before long...

re. Bob's comment: "XFN and hCard are good for what they are, but they’re still a far cry from the expressivity of FOAF.". Indeed, they also help considerably with the problem of expressing this kind of machine-readable data in HTML. But "FOAF would be a lot more useful if it didn’t rely on RDF." - the power, the expressivity of FOAF comes from RDF.

I suspect Bob (like Mad William) may be assuming RDF (including FOAF) has to mean RDF/XML. But RDF is a model, it doesn't have to serialize to XML, there are other formats - including HTML (via eRDF). Note too that microformats like XFN and hCard can be interpreted as RDF (including FOAF terms as appropriate) with GRDDL.

(Incidentally I got to this post before even looking in my personal agg via Marc Canter's on Planet Web 2.0)
Steve
2006-08-19T17:37:36

But you're forgetting that a lot of people, myself included, like to represent themselves in different ways on different sites. This would only really be useful as a starting point, I wouldn't want all my profiles to be identical. I don't use the same info for business (linkedin), school (facebook), friends from home (myspace), people i know from various music forums (last.fm) etc.
Bob Aman
2006-08-19T22:24:28

Danny, I am actually aware of all of those acronyms, however, I completely disagree on the point about FOAF's expressivity relying mainly on RDF. I would contend that FOAF's expressivity actually relies far, far more, simply on XML's ability to namespace things, and thus allow for extensions. Something that a simple XML dialect would also have, but with the added bonus of an infinitely simpler parser, fewer confused programmers, and quite a bit less bandwidth consumed.
Tantek Çelik
2006-08-20T07:22:55

Indeed Les, this is exactly what hCard solves on the Web with the minimum of duplication of data into hidden places and mimetype hackery/configuration (problems of the other approaches mentioned).

You can either use a 3rd party service like claimID as mentioned to claim your various profiles on the Web, or you can simply continue to publish your profile information on your about or author page, simply add just a bit of hCard markup, and use XFN Identity Consolidation (rel="me") to turn your about page into your consolidated identity hub.

While services will undoubtedly continue to ask you a hundred profile questions, the better ones will do something as simple as use the simple open source hKit Microformats Toolkit by Drew McLellan to let you simply enter your hCard URL and fill out all the questions automatically from that.

And the best services will help you, the user, keep your profile as portable as you would like it, by marking it up with hCard as well. Already, profiles on numerous services, in addition to the abovementioned claimID, like Flickr and Technorati are marked up with hCard and thus ready for reuse with services that use hKit.

Finally, some have brought up "extensibility" as something that is desirable yet recent lessons have taught us that in terms of standards and interoperability, the opposite is true. The more well-defined not-infinitely-extensible a standard is, the sooner implementers know they are "done" implementing it (interoperably even!), and the less chance there is of a zillion non-interoperable extensions randomly emerging (i.e. Tower of Babel problem) which then cause headaches not only for all implementers, but users as well as the users then have to contend with their data not quite working interoperably across all the different services that all support slightly different sets of incompatible/non-interoperable extensions.
Mike Macgirvin
2006-08-22T20:38:14

I agree with Steve. Who am I? It depends. Who are you, and based on that - what do I wish you to know about me? Who are my friends? Same thing. Even if I wanted to divulge them, I then need to respect their privacy wishes and decide if they want you to know that they are a friend of mine (or not) and how they came to be tagged as friends. A web service which tries to consolidate all of this info is presented with too many privacy conflicts to ever get it right (IMAO). I can recall this concept being tried by a hundred different companies (not to mention open source projects) going back to the early 90's. It was never an issue of technology. The hurdles are primarily social and revolve around trust and privacy - and delegation of these rights to third parties in language even my mother would understand.
Wohnaccessoires
2007-03-16T22:04:59

But you’re forgetting that a lot of people, myself included, like to represent themselves in different ways on different sites. This would only really be useful as a starting point, I wouldn’t want all my profiles to be identical. I don’t use the same info for business (linkedin), school (facebook), friends from home (myspace), people i know from various music forums (last.fm) etc.