Thursday, April 21, 2011

A Bend in the Ganges

I just finished reading a lovely book by Manohar Malgonkar, "A Bend in the Ganges"

The book is about two young guys around India's independence, and the turn of events in their life. The book explores interesting concepts of filial admiration, the nature of love, duty and morals. It is a very insightful read, even though the end is somewhat lukewarm. In its best places, it is a mixture of "Crime and Punishment" and "Anna Karenina".

This book was recommended to me by a close family friend, who is aware of my general interests, but has little idea of which books I read. In this case, the recommendation was spot on: I loved the book. This person is keenly aware of my general interests: he has seen me grow up over the years, though he is very unaware of my specific interests.

In general, book recommendations with such sparse information is a very difficult problem. Tying my interests together to suggest a book which has nothing in common with what I read is difficult. Many companies: Netflix, Pandora, Facebook, Google are trying to do this, and I suspect for most of them, the answer is 'social'. If Pandora knows who my friends are, they'll try suggesting songs that my friends play. Where this gets really complicated is using my preference in one domain to influence my choices in another. Using my choice of music to make intelligent book recommendations, for instance.

To make things more complicated, I weigh book recommendations differently for people: I know that I'll never be interested in a book person X recommends, but I'll always be interested in a movie the same person recommends. This is another field that 'social' might not completely answer, though there are experiments with different possible approaches.