Algo-meeting

Today I discussed with my supervisor and Patrick on the possible shape of the algorithm for the clustering of the messages. The starting point is that we need to define similarity between the messages so to allow for two kinds of interaction:

a- spatial browsing, to show the available resources;

b- spatial information retrieval, to pinpoint to relevant resources matching a query.

In both cases what we need is a mapping from the matrix of semantic distances to the matrix of the geographical distances. What we can use for this scope is the “semantic cohesion”, an home-made marker that determines the level of semantic similarities between a group of messages. Mapping this marker against the radius of the spatial cluster (r) we might find some “drop zones”, or parameter if this ‘r’ for which there is a substantial drop of similarities which will signal the existence of the cluster. Using this we wont incur in the problem of the local maxima.

There will be two levels in the support of interaction. The first will be the construction of the geo-semantic patters that will work at visualization level but not on the data level. The second step will be to support the interaction through the query/retrieval and the user’s feedback (i.e., social rating of the messages).

Tags: , ,

Leave a Reply