I feel like normalization would be a nightmare. Consider all the mistranscriptions, OCR errors, and different names in the libraries (case, parentheticals, etc).

If we assume there's no reliable way to define a book, maybe locally sensitive hashing could help find probably same books.

The idea is pretty cool though.