It'll be interesting to see if readers like the system, because I imagine there's a number of problems with the metrics and then establishing a good personal weighting function. What about books that have mixed perspectives, 1st and 3rd person? Most of the writing style metrics would be far less significant than interesting plot and good characterization. Let's say someone liked Jurassic Park. Chances are they liked it despite the slow start, yet the structure analysis apparently recommends other books partly on a criterion ("slow start followed by fast pace") that should be negatively weighted.
Pandora's music DNA matching might be more successful because listening to music seems more visceral and less cognitive than reading books. The dimension space of book recommendation feature vectors could dwarf that of music and even movies. I wonder how the system would do in the Netflix challenge by turning the analysis on the production screenplays for the films.
Pandora's matching doesn't have a very good hit rate recommending music to me. But it's okay for automatically generating a radio station, so that I don't have to be involved in selecting every song.
With books, I can't listen passively while doing something else. I'm involved in each and every book selection I make. When I'm involved, I'm already better at narrowing down my selection to things I might like than Pandora is.
Could you make the site usable by people who don't want to sign in?
Pandora had a feature where you enter one song, and they tell you what they'd recommend. That intrigued me enough to sign up. That might be useful for BookLamp too.
Pandora's music DNA matching might be more successful because listening to music seems more visceral and less cognitive than reading books. The dimension space of book recommendation feature vectors could dwarf that of music and even movies. I wonder how the system would do in the Netflix challenge by turning the analysis on the production screenplays for the films.