Machine learning sound like a scary expression. Why would we let machines learn, what would they learn and what would be the end result of such a learning experiment?
Well, forget about the "machine" word and focus on learning. Learning has a strong positive connotation. This is the very quest by which we improve ourselves, we improve our relations to others, we strive for the unknown, the seemingly unreachable. Even if we know that there will always be far more that we don't know than what we do know we continue the quest. Even the finest scholars spend their life learning.
Now we have machines and chief among them computers. They are the vehicles through which we browse the web. But browsing is not enough. More often than not it boils down to a quasi random promenade in an ocean of content. To use a well-known quote, we drown in content but starve for knowledge. This is for instance why we have librarians who, through their skills, try to save us from drowning. Musing in a physical library is highly structured, organized. Books sit on shelves, not any shelves though. These are the shelves that librarian classification have elaborated over years of efforts sorting out books. In a sense, a physical library is a tribute to Euclidean geometry whose motto is:
Physical library = Books + Readers
The physical space is structured such that the books affordance is maximized for patrons walking along shelves.
Musing in a digital library is very different. A digital library is a tribute to non-Euclidean geometry and it goes by the following motto:
Digital library = Information
This is this very high-dimensional information, our footprints that we leave behind us (what we read, what we have in our booksshelves, how we organize our bookshelves, what keywords we type in, what tags we choose and so on and so forth), that machines can learn to tame and to decipher. This requires a fairly intensive dose of (non-Euclidean) geometric modeling and the appropriate treatment of large set of data that go well beyond the classical (flat) Amazon recommendation system.
Now if you are willing to take the scientific gamble and struggle with data, matrices, eigenvectors and the whole shabam, you will not be disappointed and this is precisely what we have done. This will be released soon. In the meantime, here is an appetizer: This is what we call computing books paths or books musings or book promenades... These are not random promenades. They are carefully computed based on what the users of the digital library have done.
They used to say at the French Railway, "beware, a train may hide another one", urging you to prudence. Here we say "guess what, a book may hide several other books" and we advise you to follow the guide! A picture says it all:
Fig.1: Picture drawn from Cyberlibris library algorithm.
Here the cube contains the whole library which means that, on an iPad, you will hold the whole library in your hands, at your fingertip. Not any kind of library though. This is how, given the library users habits, the library should be structured. Now, based on this, one can trace from any single book promenades that will help you discover other books that you would have had more difficulties discovering otherwise.This is most useful for readers, authors and publishers. At least, we try to make sure that no single book remains orphan for too long. The qualitative dividends of this quantitative effort are manyfold. Promenades are one of them. And, by the way, at the risk of playing to hard with words, manyfold is possible because we view the digital library as book manifolds.
In a nutshell, united digital library users are stronger than isolated users and this strength is delivered by... machine learning!