[Sigia-l] Semantically linked library catalogues

[Yogesh Tadwalkar]

Hi,

For a library project here in Singapore, I am proposing semantically linked
catalogue of books. I am writing this post to solicit ideas about how this
can be best accomplished. Sorry for a long post.

In simple words, the idea of semantic linking is that books in the library
will be indexed and linked together NOT just based on topics or other
traditional methods but on more granular things like content of the books,
etc.

For example, although 'The Fountainhead' by Ayn Rand and 'A Pattern
Language' by Christopher Alexander are not traditionally linked because one
is a novel and the other a semi-handbook; they are indeed semantically
linked because both talk about architecture and architecture for use - not
just decoration. Similarly the book on 'Sun Tzu' by Samuel Griffith,
although is about Chinese art of war and leadership; is semantically linked
with books on the Japanese martial art of 'Akido' because they both believe
not in pure aggression but rather weakening the opponent by using his own
energies and 'winning without conflict'.

The simple idea here is that although such semantic linking is part of many
human conversations between friends, a pupil and a teacher; it has not yet
been well captured in a traditional library catalogue. The idea is to help a
reader to know what books to read next based on more qualitative
recommendations.

This is different than 'People who bought this book also bought...' by
Amazon because Amazon:

1. Does this based mainly on topics and not granular things like content
similarities. I have not seen a book on electronics listed in 'People who
bought...' with a book on usability yet.

2. Mainly uses collaborative filtering based on buying and rating behavior.
That means a user's preference or recommendation is not captured until he
has bought something (I believe that is how they work. or is there something
else?)

What do you think is the best way to create such semantic linking?
Interviews, questionnaires, surveys, with readers? Or some modified form of
collaborative filtering? How do we capture the qualitative human
recommendations best and convert that into a semantically linked electronic
catalogue?

Once again, sorry for such a long post. I welcome your ideas.

regards,
Yogesh Tadwalkar