Rough Clustering Generated by Correlation Clustering

Correlation clustering relies on a relation of similarity (and the generated cost function). If the similarity relation is a tolerance relation, then not only one optimal partition may exist: an object can be approximated (from lower and upper side) with the help of clusters containing the given object and belonging to different partitions. In practical cases there is no way to take into consideration all optimal partitions. The authors give an algorithm which produces near optimal partitions and can be used in practical cases (to avoid the combinatorial explosion). From the practical point of view it is very important, that the system of sets appearing as lower or upper approximations of objects can be taken as a system of base sets of general (partial) approximation spaces.