I’ve been meaning to take a look at BookLamp for awhile, a site that describes itself as being like Pandora for books. Where LibraryThing helps readers find books based on similarity of taste, BookLamp takes a more technical approach, scanning the books and, well, here it is in their words:
So the first thing that we did was that we scanned a book. We built a program that automatically breaks that book up into scenes. From there, we had the program go through and identify all the different words in the scene. So adjectives, nouns, adverbs, comma usage, things of that sort. Then we sat down and we searched that information for patterns. English is a very self-descriptive language. Adjectives describe things. Verbs are action. And I believed, that by looking at the types of words that made up a scene, you could make some educated guesses about what kind of content was in the scene, without a human ever actually reading it.
BookLamp analyzes text for the ways that density, pacing, action, dialog, and description can be mathematically expressed, then creates charts for the books and looks for books with similar charts. Despite the claim that “there is no doubt that we’re tracking stylistic consistencies between authors,” I have to wonder if it would be possible to game the system by adding the works of authors who use prose in truly unconventional ways, like Beckett, Burroughs, and Borges.
To be fair, even human readers’ advisors can struggle with unconventional works. But the skeptic in me imagines that a computer program that focuses on formulas might miss some of the harder-to-express nuances that make recommendations good. And what is the ultimate goal? To give readers books that deliver the exact same reading experience each time? Or to broaden readers’ experience by giving them books that have some elements they like and some that are new to them? Although, again, that might happen with BookLamp, too: two books with the same arcs of action might have completely different characters and settings.
Right now, BookLamp is still in the beta stage. It has only analyzed 179 books, all of them sf or fantasy, so it’s hard to get a sense of how well this will work on a wider scale. It would be fun to see what non-sf recommendations it might make for a sf title, or vice versa, or to see what read-alikes it suggests for Beckett, Burroughs, and Borges.
What do you think? Would this be a fun thing to play with, like Pandora? Or does the phrase “without a human ever actually reading it” chill your bones?
Watch this video for an overview: