LibraryThing has added a few different statistical recommendation lists (mine). This is really fun to play with.

I use very general tags (small library) so the by-tag view says basically nothing. I don’t seem to have enough non-fiction to get interesting results there (most of the entries are pretty clearly based on just one or two of mine). The interesting stuff is going on in the “people with your books also have (fiction)” category.

  • I’m pretty clearly out in the genre zone. I count almost exactly half as sf/fantasy/horror — although the list moves from genre-heavy to classic-lit from top to bottom.
  • A big chunk of those get dropped by the “omit authors you own” switch. (Genre fans tend towards complete collections?) The proportion drops to 40%, the lit heads the list with the genre stuff further down, and there are more entries that belong in both camps (Atwood, Huxley, Orwell).
  • The recommendations are generally right on the button: lots of books I own in New Zealand, many more that I’ve enjoyed but never got around to buying.

That last point is a problem though: in the unfiltered list there are no authors I’ve never heard of, and only a few I haven’t read. In the filtered list they turn up within the first ten, but the first unfamiliar name is down below 50 (a grand total of four author/title pairs didn’t ring any bells). I think what’s going on is that my collection is too small, so the results are getting heavily influenced by overall popularity; lots of the same books turn up on the “most popular you don’t own” list. I’ve started adding wishlist-tagged books (instead of storing them on Amazon, a site whose design I detest but which is terribly convenient), to give the system a longer baseline.

Finally, a couple of pleasant surprises: The Secret Garden (via The Weirdstone of Brisingamen, I guess?) and The Princess Bride, I don’t know where that comes from but I’m flattered. But the best is Twisty Little Passages: there’s nothing overtly on game design or interactive fiction at all in my lists, although it’s definitely one of my interests, so this is clustering doing its job perfectly. Impressive.

Update: Ok, now I’ve got the Inform Designer’s Manual, which I downloaded, printed, and bound but hadn’t thought to add to LT. But the recommendation came before, I swear!