I recently read Chris Meadows' assessment of Booklamp, the online face of the Book Genome Project. Meadows has profiled the book-discovery algorithm in the past, and he returned to review what had changed.
Meadows finds two things are missing from the latest version of Booklamp: selection, and humor. The latter reflects development still needed in the matching function, while the former is a byproduct of limited access to content, a problem often faced by startups who want to work with publishers.
One way or another, discovery is a problem publishers pay to solve. In May, Perseus even funded efforts to 'hack' book discovery, ultimately awarding a $10,000 prize to Evoke for a platform that generates "character maps" to compare different works.
Given the importance of discovery to publishers, it confounds me to hear (again; cf Small Demons) that a service designed to improve discovery struggles to get publishers to join in. Online, the ability to discover content depends on access to content. Maybe 2014 will bring a new refrain.