Graph-Based Clustering with Constraints

A common way to add background knowledge to the clustering algorithms is by adding constraints. Though there had been some algorithms that incorporate constraints into the clustering process, not much focus was given to the topic of graph-based clustering with constraints. In this paper, we propose a constrained graph-based clustering method and argue that adding constraints in distance function before graph partitioning will lead to better results. We also specify a novel approach for adding constraints by introducing the distance limit criteria. We will also examine how our new distance limit approach performs in comparison to earlier approaches of using fixed distance measure for constraints. The proposed approach and its variants are evaluated on UCI datasets and compared with the other constrained-clustering algorithms which embed constraints in a similar fashion.