Marcel Caraciolo covers his slides from keynotes at VII PythonBrasil, the most interesting for topic mappers would be Crab – A Python Framework for Building Recommender Systems.
Recommender systems by necessity have to identify the interests of a user (2 subjects, interests and user), match those to other interests (another subject) and then produce a recommendation (yet another subject), plus relationship subjects if you are interested. Recommender systems are already identifying all those subjects and gathering instances of them together.
What would you do to make their constantly interim results available to other systems?