Nice discussion today with my supervisor. We came out with this idea of building an algorithm that is able to find patterns in zones of the map. The idea is that messages are distributed on the map. The challenge is to find a system to define clusters among them based on their similarity [d1] or dissimilarity [d2]. To do it we can use several technique which avoid going to the edge of natural language processing and in particular: (1) spatial clustering; (2) social network analysis; (3) threads analysis.

Using concurrently these methods, we should be able to zone the maps and assign to each zone a set of meta-descriptors. Finally, messages in the same zone should have their content parsed by a Latent Semantic Analysis engine (4) which should generate “GeoSemantic Patterns”.

This outcome can be then feed into the search engine of the interface and to the spatial clustering engine to resolve conflicting zones.

Next challenge is then to find a good literature chunk for each of those topics.