This Project's

Nootropia: Adaptive Multi-topic Information Filtering

Theme: Natural Language Processing
In Information Filtering, unwanted documents are filtered out on the basis of their (dis)similarity with a user profile. Many models exist, but most suffer from two draw backs. Frequency based profiles cannot represent more than one topic for filtering, so users with multiple interests must initate multiple profiles. As a side-effect, adpatation of profiles, as user interests evolve, is problematic. Nootropia solves both these problems. Using self-organising term dependency networks, it is able to represent multiple topics in a single profile, that is able to adapt to both short-term and radical changes in a user's interests. It can grow new topics, and forget old ones. We have shown that Nootropia's adaptive properties display the characteristics of auto-immune inspired systems.