Topological Collaborative Clustering
Nistor Grozavu, Younès Bennani · International Conference on Intelligent Information Processing · 2010
This work presents a collaborative clustering approach, allowing to take into account another classifications results without recourse to the data in an unsupervised learning context. The approach is presented in the case of Kohonen Self-Organizing Maps and valid for all prototypes based classifications. Having a collection of distributed datasets on several different sites, the problem is to cluster each of these datasets by considering only the local data and the distant classifications from other collaborative datasets, without sharing the data among different centers (datasets). For this purpose, our approach is divided into two steps: a local and a collaboration step. The local step uses the classical algorithm of Kohonen, local and independently on each of datasets, which will result in obtaining a SOM map (Self Organizing Map) for each of those bases. The collaboration step would be collaborating each dataset with all the SOM maps associated with other databases obtained from the local step. The article presents the formalism of the approach and its validation.