Quarter Life Crisis

The world according to Sven-S. Porst

« ☃MainTrifle »


445 words

Paul recently bookmarked a text discussing how useful tags are which goes on and on in a pseudo-scientific analysis to conclude that they’re mostly pointless. Paul then commented:

An interesting, somewhat long, survey into tagging, and how much it contributes over titles and descriptions. “How come the same person who tags his photo {Milan, Italy} is able to title it Galleria Vittorio Emanuele II?” The answer to this, I suspect, is that uploaders encourage batch tags but not batch titling. Also, as I noticed in where? what? when?, top tags for places tend to be geographic, but not terribly specific. I’d love to see a well-thought-out riposte, mind.

I don’t really think a riposte is possible here. Isn’t the text just stating the obvious: tagging doesn’t work?

And how should it? If thousands of imbeciles (aka ‘the public’) throw random words at objects, it’s unlikely that the results you get out of those tags is any better than a general Google search on their title and comment should be.

Particularly so as what they call ‘tagging’ on the net today is so dumb and simple-minded that it hurts.

While I am pretty sure that it will be rather imperfect, watching the iPhoto ‘09 ads looked promising in terms of tagging as they seem to be smart enough - or at least attempt - to use context as well: Paris is in France etc. This is completely missing in today’s ‘web 2’ tagging systems. As are synonyms &c.

I guess my point ends up being something like this: There are two models which work reasonably well: The old school librarian model with a well defined taxonomy and educated people systematically ‘tagging’ things and the new world Google model with a very egalitarian approach and loads of statistics thrown in to extract the relevant keywords from any kind of text.

Somehow tagging tries to avoid the good parts of both of these approaches (educated people, clever statistics) and just gives you dumb tags. That always looked like a missed opportunity to me. Or rather it looks like one of those typical geek things which are easy to implement with a little SQL and no thinking.

And I think Paul is spot-on about the usability aspect: Frequently the user interfaces make batch tagging is simpler than batch renaming. Hence it is used more frequently. Likewise, tags are sometimes actually useful because they give you a simple way of finding things by offering single-click searches for all items with the same tag. But I think that a good implementation of general search should give equally good results. It’s just that sites which use tags tend to have sucky search.

January 13, 2009, 22:04

Tagged as tagging, web, web2.


Comment by ssp: User icon

In true style, Paul comments by placing a link to this page with his comment in delicious which I replicate here so it doesn’t escape those without X-Ray eyes.

I suppose one could have a full ‘discussion’ that way. I’m sure some geek is working on an aggregator already, but I’ll just go for convenience this time around.

January 14, 2009, 10:28

Comment by g: User icon

I find tagging useful for myself, I tag more often on del.icio.us but it’s for me and over there my most abundant tag is: http://delicious.com/g/system:unfiled. It’s about 25% of my bookmarks, and that’s probably because I couldn’t think of a way to class the link without writing a long narrative with lots of blah_blah_blah…

This has been said before and I think it’s important. I noticed that normal people who use tags tend to write sentences like, "war in gaza" in the tag field and end up with "war" "in" "gaza" as three separate tags. A system that relies on a person to – continually – think about the limitations is weak. Now that's where semantics enter the tag discussion, which I’m sure the tagophiles hate. A search for "war" would yield a lot of results, fine. Would a search for "war in gaza" give uniquely relevant stuff without wading through acres of crap? It seems the system was designed for machine readable people, and one has to adapt to think in the same manner. Personally, when I write and talk I don’t say “hyphen” and “underscore” in between each word, or contractmysentenceintooneword.

January 14, 2009, 11:30

Comment by g: User icon

Oh dear, I completely forgot about the Markdown filter.

January 14, 2009, 17:04

Comment by ssp: User icon

Let’s say I’ve seen worse and corrected things as I saw fit. But we surely could open a ‘Victims of Markdown’ club… Oddly I always find the Markdown documentation abysmal. Whenever I need to look something up there the stuff I’m looking for is really well hidden while the obvious stuff is all over the place.

January 14, 2009, 18:01

Comment by Mark Bernstein: User icon

I think you’re mistaken when you dismiss Dr. Marshall’s study as pseudoscientific. The methodology is unconventional, to be sure, but to my knowledge no known methodology addresses this question as well.

One could, of course, create a test case with a set of known images, each tagged by purportedly typical or representative users — typically undergraduate students. This assures that we know the entire universe of targets, and we can thus measure precision and recall conventionally. But are our tags representative of actual use by actual users? This is difficult to argue, and in general cannot be determined.

In Marshall’s study, we can only estimate precision and recall by assuming that she has, in fact, found all or nearly all the pertinent images (or, alternatively, estimating the number she has missed). This is problematic, to be sure. But we do have the benefit of actual tags created in the course of actual use, for whatever purpose actual users have in mind when they create tags.

For whatever weight it carries, Dr. Marshall is a senior researcher at Microsoft’s Silicon Valley lab, which she joined after many years at the legendary Xerox PARC. She’s twice won the Engelbart Prize. She’s regularly on program committees for JCDL, HT, CHI, WWW. She’s got a list of scientific publications as long as your arm.

January 18, 2009, 14:31

Comment by ssp: User icon

Rest assured that I neither doubt nor know about Mrs Marshall’s proficiency at her profession.

All I saw – spelled out in exhaustive detail on the web page I linked to – looked like an enthusiastic ‘web 2.0’ user spending a lot of time digging into the photos available on flickr, taking samples on a single location and counting the numbers.

All this for a result which one should reach a priori from the conceptual weaknesses of how tagging is used and implemented in ‘web 2.0’ sites today.

January 18, 2009, 16:00

Comment by ssp: User icon

Interesting further comment on the topic: Tags do work (for me, at least) which throws tags on delicious into the comparison.

Personally I think that tagging on delicious is bad as well as it suffers from the same conceptual flaws as flickr’s. But they have a few tweaks like tag autocompletion and suggestion of common tags for a link (is that a good or a bad thing? I gravitate towards the latter…) as well as tag grouping if a user is inclined to spend a lot of time on the site.

January 19, 2009, 11:04

Add your comment

« ☃MainTrifle »

Comments on




This page

Out & About

pinboard Links


Received data seems to be invalid. The wanted file does probably not exist or the guys at last.fm changed something.