Clustering ensemble selection methods choose qualified and diverse base clusterings for ensemble. Existing methods rank base clusterings according to validity indices as quality measures and select diverse clusterings in top qualified ones. However, the ranking-based selection is hard to filter out base clusterings and may miss important diverse clusterings of moderate quality for ensemble. Aiming at the problem, we revisit the base clustering selection from the view of stochastic sampling and propose a Clustering Ensemble Selection method with Determinan-tal Point Processes (DPPCES). DPP sampling of base clusterings adds the randomness to the clustering selection while guaranteeing quality and diversity. The randomness is helpful to avoid the local optimal solution and provide a flexible way to select qualified and diverse base clusterings for ensemble. Experimental results verify the effectiveness of the proposed DPP-based clustering ensemble selection method.
展开▼
