Constrained clustering with a complex cluster structure

摘要: In this contribution we present a novel constrained clustering method, Constrained with complex cluster structure (C4s), which incorporates equivalence constraints, both positive and negative, as the background information. C4s is capable of discovering groups arbitrary structure, e.g. multi-modal distribution, since at initial stage classes elements generated by constraints are split into smaller parts. This provides detailed description elements, in relation. order to enable an automatic detection number groups, cross-entropy applied for each partitioning process. Experiments show that proposed method achieves significantly better results than previous approaches. The advantage our algorithm increases when focusing on finding partitions clusters.