April 9, 2006
Tim Spalding of LibraryThing (the plural version of which appears to be a squattor) recently wrote a nice (and funny) little response to a multi-versus-single-word-tag thread on the TagDB mailing list.
For good or ill, most web users have no such intuition. 99.9% would never speak of the “union” of two tags, this being some sort of trickle-down from set-theory talk. Far fewer would have that intuition in terms like “find all pictures tagged both ‘london’ and ‘trip.'” And of those, few would have any idea how to do it. Yes, most search engines allow all sorts of clever boolean logic (+london +tip -“pigeons shitting on me in trafalgar square.”) No, nobody uses that logic.
“London” and “trip” make sense on their own; “Spring semester” does not. The union of “spring” and “semester”? How about the union of “spring,” “training” “red” and “sox”? Or shall we look for the union of “springTraining” and “red_sox”? Congratulations, you need an “about” page to tell people how to tag, and your users are all programmers.
This conversation is necessary among any team implementing tagging. It’s a basic usability point.
In the single-word corner, we have Ihe tag purists. To them, tags are unstemmed tokens, and tokens are atomic. If you allow phrases and even sentences (e.g., “my trip to london”), you’re moving beyond tags/reverting back to captions/keywords. Tag purists also view tags as facets of an item–attributes or metadata describing some aspect of the item–and frown on tags that seem to represent (or replace) content.
In the multi-word corner, we have what I’ll (lovingly) call the usability police. To them, as to Tim, any notion of the “intuition” of using lists of single words to file and then retrieve items is absurd: most users have no such intuition. Forcing users into fusing words together (e.g., “red_sox”, “hot_dog”, “beer_belly”) is just asking to limit your reach to only the most technically inclined.
The tag purist perspective makes a lot of sense from the IR point of view. Computing tag intersections and unions, related tags, tag clusters, etc., is much simpler than doing the same with collections of full-text documents. This is appealing. From here, it looks as if tags basically distill the essence of a thing down to a few words. It’s like calculating statistically relevant words for free. Plus, it sort of pushes off the question of “Do I use a real search engine, or do I just go with the standard three-table MySQL tag schema?” until your system really, really needs to scale, since there’s no need to compute or analyze anything.
It also makes serving your customers’ needs easier. If I’m looking for things tagged “london,” I’d expect to get back things about people’s trips to London, or London bombings–which won’t happen if another user tags her photos/blogs with “London Bombing” or “Trip to London” or “London – March 2006” without some back-end analysis. But if you do that, then you lose the atomicity of a tag like “hot dog.”
The compromise position some sites have reached is to push multi-word tags inside of quotation marks. That poses a bit of a parsing issue, though (not for the code, but for the human trying to maintain/scan the list).
I think part of the anxiety people are having over multi- versus single- word tags is due to the early hype surrounding tagging as a cure-all for site navigation. In truth, any user input requires the application of IR concepts. Just as full-text indexing results in ambiguities, false positives, and false negatives, so too will any tagging scheme, no matter how intuitive to its users.
The advantages that I see for tagging are:
- it provides an obvious navigation alternative,
- it allows for dynamically expanding facets (sets of attributes) of items, and pivoting on each,
- it’s simpler to index and cluster around a subset of words that the author/reader feel really represent an item, than it is to try to automatically distill the essence of the same item from a collection of hundreds of words,
- it reduces the effort required to input metadata for visual items, and
- it increases the opportunities for tools to automatically add important facets to an item.
As for my own opinion, I lean heavily toward multi-word tags separated by commas when the only possible input of tags is through a single textbox. I believe, however, this will be rendered moot in the next year as new graphical–and usable–tag management interfaces are developed (and I don’t mean a series of checkboxes on the right-hand side of the page). But this is going to force people to think in terms of IR concepts who’d rather not.
(Dodge an issue? Why, I’d never!)