Скачать или смотреть Constrained Clustering via Post-processing

Title: Constrained Clustering via Post-processing
Authors: Nguyen-Viet-Dung Nghiem, Christel Vrain, Thi-Bich-Hanh Dao, and Ian Davidson
Abstract: Constrained clustering has received much attention since its inception as the ability to add weak supervision into clustering has many uses. Most existing work is algorithm-specific, limited to simple together and apart constraints and does not attempt to satisfy all constraints. This limits applications including where satisfying all constraints is required such as fairness. In this work, we take the novel direction of post-processing the results of a clustering algorithm (constrained or unconstrained) as a combinatorial optimization problem to find the best allocation of instances to clusters whilst enforcing constraints. Experiments show that when evaluated on a ground truth, our method is competitive in terms of clustering quality with the more recent approaches while being more computationally efficient. Finally, since all constraints are satisfied, our work can be applied to areas such as fairness including both group level and individual level fairness.


23rd International Conference on Discovery Science
Webpage: https://ds2020.csd.auth.gr/​
Proceedings: https://www.springer.com/gp/book/9783...