Cohesion and Repulsion in Bayesian Distance Clustering

Clustering in high-dimensions poses many statistical challenges. While traditional distance-based clustering methods are computationally feasible, they lack probabilistic interpretation and rely on heuristics for estimation of the number clusters. On other hand, model-based techniques often fail to scale devising algorithms that able effectively explore posterior space is an open problem. Based recent developments Bayesian clustering, we propose a hybrid solution entails defining likelihood pairwise distances between observations. The novelty approach consists including both cohesion repulsion terms likelihood, which allows cluster identifiability. This implies clusters composed objects have small "dissimilarities" among themselves (cohesion) similar dissimilarities observations (repulsion). We show how this modelling strategy has interesting connection with existing proposals literature as well decision-theoretic interpretation. proposed method efficient applicable wide variety scenarios. demonstrate simulation study application digital numismatics.