Wednesday May 16, 2007

LibraryThings tags added to a public library catalog - some thoughts

OK, time to try to breathe some new life into this blog

Yesterday I read about LibraryThing's tagged data being included in the Danbury, CT Public Library's catalog. First, bravo to the Danbury Library for being gutsy enough to go first and experiment with this.

From what I've read of the commentary on NGC4Lib and elsewhere, the results aren't perfect, but they are pretty good. And hey, the results weren't perfect before from just the LC headings, so I believe that mostly this will help. I sometimes wonder, when people point out counterexamples for how searching fails in some particular case, did they actually expect it to work well always?

The emails tend to point out some weirdness in recommendations based on LT's social data, but that's exactly the crux of social data. Odd connections and preferences emerge, especially if the title has a small set of tag data - whether they are in fact helpful is harder to tell because helpfulness in the this context is bound to be so personal.

Years ago I worked part time at a public library as an assistant and one of the most challenging questions I was regularly asked was, "can you recommend a good book?"  I could tell you several, but the odds are unless you read and are exactly interested in very similar things as me, you may not like my recommendations. Or maybe you will, who knows? Being able to do that well was an art and it took time to really begin to know authors, genres, reading habits, etc and not focus on my own personal tastes. It wasn't easy and all this was before any social networking tags let alone Amazon reviews.

I looked at the Danbury catalog yesterday and put in the title I'm currently reading, Our Mutual Friend, by Charles Dickens. I guess LT's data worked well enough; several other Victorian novels recommended; though only one by Dickens weirdly enough. And the tags themselves made sense. But Dickens has a lot of tags - he's one of the top 20 LT authors by copies and even though OMF is not one of the most famous of Dickens' works, LT still has data on 687 copies (I salute all of those people for taking the time to tag this good book). When I actually clicked on one of the tags and went into the tag browser, I got a really nice reading list of 19th century fiction, all of which were at that library. Same thing for the tag, "Dickens."  And for "London." Nice lists, not random. It was easy to use as a browsing tool and something I think I would enjoy as a patron.

A lot of the power of LT and tags comes when there are enough of them for any given title or author to statistically overwhelm the odd tags and entries that are too idiosyncratic to be of much use to anyone except the tagger herself. I'm not sure how helpful this would be for a really obscure title that LT has little data on - so perhaps I'm wondering just how good tagging is at reasonably producing Long Tail recommendations. Tagging is something that, like Amazon reviews for a given title, while perhaps possessing a Long Tail distribution,  I believe only really displays its power in the much smaller set (of titles, of reviews that are tagged as helpful) that accounts for most of the data.

My conclusion after that last run-on Dickensian sentence is that adding LT's huge tag dataset could be a very interesting addition for most public libraries, especially given the price which is supposed to be very low. What's to lose?