Latent Spatial Dirichlet Allocation

Junsouk Choi, Veerabhadran Baladandayuthapani, Jian Kang

Keywords: Gaussian process, Latent Dirichlet allocation, Multiplex imaging, Spatial data analysis

TL;DR: We propose the latent spatial Dirichlet allocation, a general topic modeling framework that can be applied or easily extended to a broad range of spatial data classes and dependencies.

Abstract: We propose a novel topic modeling approach, latent spatial Dirichlet allocation (LSDA), which generalizes the latent Dirichlet allocation to spatial data. LSDA integrates spatial Gaussian processes within the LDA framework, thereby effectively capturing complex spatial dependencies inherent in spatial data. We develop an efficient Markov chain Monte Carlo algorithm, and applications to both real and synthetic datasets successfully demonstrate the utility of LSDA.

Submission Number: 96