Today I discussed briefly the algorithm for my thesis with my advisor. We decided to not spend to much time trying to support multi-languages analysis in the system. Then we decided also that a better way to go is to use semantic distance an heuristic for spatial clustering. This basically brings me to a choice:
1. I find a way to translate a matrix of semantic distances into some form of values to pair with the geographical two-dimensional space. In this case I can proceed to checkout with an available clustering method with non-geographcal features.
2. I find a way to “parametrize” the clustering using the semantic distances… not clear how.
3. Compute with the semantic distances a semantic coherence that is the output of the spatial cluster. Then running many times the spatial clustering to maximize the outcome of this coherence.