Constraint projection for semi-supervised cluster ensemble

Semi-supervised cluster ensemble takes the advantages of ensemble learning and semi-supervised learning. In this paper, constraint (cannot-link and must-link) projections are illustrated for semi-supervised cluster ensemble (CPSCE), a hierarchical semi-supervised cluster ensemble algorithm. It is flexible for the relaxation of some constraints during the learning stage. First, the data points of instance-level constraints and base clustering ensemble results are together projected in a lower dimensional space guided by the constraints. Then, mesh partition software (METIS) is performed on the similarity matrix. Finally, a few datasets are chosen for experimentation from the UCI machine learning repository. The results show that CPSCE performs better than some existing algorithms. © 2015 ICIC International.