US Patent 7870136 Clustering data with constraints

A method for clustering data using pairwise constraints that includes receiving a set of data for clustering, the set of data includes a plurality of data units; identifying soft pairwise constraints, each indicating a relationship between two of the plurality of data units in the set of data and having an associated confidence level indicating a probability that each pairwise constraint is present; and clustering the plurality of data units in the set of data into a plurality of data partitions based at least on a chunklet modeling technique that employs the soft pairwise constraints.