It’s really helpful for me to think more deeply about tagging than I usually do. I do tag in del.icio.us and in LibraryThing (though I will confess that I am so trained to think in controlled vocabularies that I actually look up the LCSH for my books on LibraryThing and use those as tags, along with the Dewey class number…).

But I mostly don’t see the big benefit. I tag in LibraryThing so that I can find titles quickly. Still, my major use of LibraryThing is “what books did I move when I moved to NYC”.

So, I’ve nothing useful to add, but just wanted to say I found the post and the comments interesting, thought-provoking, and helpful.

Would love that evidence… I am aware of it myself from testing at LII (when people *do* use advanced search, they get it wrong, too).

I will try to dig up some evidence for you on “people-don’t-use-advanced-search” because they really truly don’t use it.

I have thought a lot about tagging. Both in our industry, and in my past one — legal marketing — where lawyers do tons of “workflow tagging,” except they call it “billing and timekeeping.”

Part of the problem with how tagging is considered for things like books is that we tend to think of it as a synonym for “bookmarking” in the Web browser sense of the word because of a Web-centric view of services like del.icio.us and Google Bookmarks.

When I go back and review my experience with legal billing, the good software products allow you to set up a system with workflow at the center of the activity; you define variables such as current clients, matters, co-workers, expense buckets, etc., with the goal of “tagging” each 1/10th of an hour as belonging to a specific set of cases, sub-cases, industries, clients, type of law being practiced, etc. In the case of this kind of billing, the point is entirely (for almost all firms) to produce a coherent “backwards view” of lawyer activity, such that billing can occur, and lawyers’ time assessed. Some few firms mine this extraordinary pile of data for CRM and matter analysis, but it’s rare.

This software ends up being able to generate all kinds of reports based on the variables that go in. And if you shift your brain about 10-degrees, you can think of the variable content as “tags,” and work backwards to ways in which you could use standard Web-type tagging to visualize the data; a tag cloud of clients in the 90-day past due bucket, for example, might be fun…

This is a type of tagging I’d call “verb” or “activity” tagging. It’s very useful, because you only tag stuff based on what you did with it, not what you thought of it, which would be “adjective” or “descriptive” tagging. Verbs are, of course, much more useful than adjectives.

In the realm of book or material tagging we’re talking about, we do see some tags like “to read” or “have read” or “I own” or “I want,” which are all verb tags; they qualify an item based on an activity; purchase, reading, coveting, etc. In the discussions I’ve seen in libraryland, these tags are commonly regarded as, at best, confusing to the overall process and, at worst, to be discarded in the public view of tags. This is because there is no back-data on the user related to overall preferences, “life-flow” or self information.

Go back to the lawyer billing application for a second… once an attorney has self-identified as a biller on a project, any hour that gets tagged to that account and lawyer is then cross-referenceable (?) to other attorney data; billing rate, practice group, upcoming vacation time, etc.

In a library setting, an initial set of queries about how a user will be utilizing tags could provide a set of data that can then be cross referenced back for his/her own use, and that of others. In my head, I call this “self tagging.” It does hinge on either, A) a user being ok with the system tracking personal usage information, or; B) an anonymizing system, or; C) both

So, for example, when I start using the system as a student, I could enter data about my major, my school, my intended career choice, current classes, professors, etc. Then, when I get to a book/resource in the system and choose the tagging option, I can tag it with what class it’s being used for from an easy drop-down.

This is useful from a workflow standpoint (what books do I need for Chem 207?) and from a discovery standpoint (what books are being used for chemistry courses at MIT?). Combine this one piece of data with a quick rating option (1-5 stars, whatever), and you’ve got a way to rank the use of text-books by number of tags and popularity. You could also compare this info for students vs. professors vs. authors vs. librarians. Wouldn’t that be fun information to have as I plan my next course?

The use of tags for ontological discovery is great, and services like LibraryThing do a good job of aggregating them. It is cool to be able to see a list of items that people I know, and whose opinion I trust, have tagged “steampunk” etc. It’s helpful when trying to find items based on “adjective” descriptions.

Workflow or “verb” tags, however, can (I think) provide a much better view of what’s actually happening with a resource. And activity monitoring is always a great teacher. So, in the example you give, the recommendations might come based less on the number of people who have tagged a work “basket weaving,” than the number of people who have tagged a work “purchased” for a “basket weaving course” with a 4 or 5 star rating.

For non-academic use, you could have a user profile with sub-domains based on use. For example, most of the books I purchase/read would fall into 10-or-less categories. So, if I self-tag my job as “marketing” and tag my industries of interest as “libraries” and “not-for-profit,” then when I mark a book as “purchased” and “work related,” it would automatically be tagged for my job/industry. If I then add one ontological tag — let’s say, “Web design” — you end up having a much richer data set every time I add something to my “purchased” set. Other folks could look for “design” books tagged to (rather than tagged “as”) marketing people, library people, not profit people, etc.

Not sure on the technical side how you’d implement something like that… I’m not a code guy. But as a user, a teacher and a data-hungry marketing guy… the idea of being able to see which resources get “hit” in various contexts would be fascinating to me.

I think that’s what Jean and I were talking about — good recommendation behavior. In other words, recommendations have to *work.* A good recommendation system knows this. Jean’s point is that Netflix was annoying when its system didn’t work, and that it’s pleasurable now that it works well.

Jonathan, your comment is also your blog post, which is fine. (It’s not pron and it’s not ad hominem, so let it bloom!) But note that one reason I tag on delicious is that it’s easy to tag. There are times when it’s not easy enough — when I’m hurrying — and then I count on the abstract working for me (since it’s so easy to add an abstract). Part of my tagging behavior is about findability, but part of it has to do with being able to do something well in the first place.

As I commented on your blog, WorldCat’s social tools are at best perfunctory when they exist at all. Look at the review function.