Bayesian Agglomerative Clustering with Coalescents

TL;DR

This paper presents a Bayesian hierarchical clustering method using Kingman's coalescent prior, with new inference algorithms that outperform existing methods in document and linguistic data clustering.

Contribution

It introduces a novel Bayesian model for hierarchical clustering based on coalescent processes and develops efficient greedy and Monte Carlo inference algorithms.

Findings

Our algorithms outperform existing clustering methods.

Effective in document clustering tasks.

Demonstrated applicability in phylolinguistics.

Abstract

We introduce a new Bayesian model for hierarchical clustering based on a prior over trees called Kingman's coalescent. We develop novel greedy and sequential Monte Carlo inferences which operate in a bottom-up agglomerative fashion. We show experimentally the superiority of our algorithms over others, and demonstrate our approach in document clustering and phylolinguistics.