An Online Gibbs Sampler Algorithm for Hierarchical Dirichlet Processes Prior
Abstract
The hierarchical Dirichlet processes (HDP) is a Bayesian nonparametric model that provides a flexible mixed-membership to documents. In this paper, we develop a novel mini-batch online Gibbs sampler algorithm for the HDP which can be easily applied to massive and streaming data. For this purpose, a new prior process so called the generalized hierarchical Dirichlet processes (gHDP) is proposed. The gHDP is an extension of the standard HDP where some prespecified topics can be included in the top-level Dirichlet process. By analyzing various datasets, we show that the proposed mini-batch online Gibbs sampler algorithm performs significantly better than the online variational algorithm for the HDP.
Cite
Text
Kim et al. "An Online Gibbs Sampler Algorithm for Hierarchical Dirichlet Processes Prior." European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, 2016. doi:10.1007/978-3-319-46128-1_32Markdown
[Kim et al. "An Online Gibbs Sampler Algorithm for Hierarchical Dirichlet Processes Prior." European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, 2016.](https://mlanthology.org/ecmlpkdd/2016/kim2016ecmlpkdd-online/) doi:10.1007/978-3-319-46128-1_32BibTeX
@inproceedings{kim2016ecmlpkdd-online,
title = {{An Online Gibbs Sampler Algorithm for Hierarchical Dirichlet Processes Prior}},
author = {Kim, Yongdai and Chae, Minwoo and Jeong, Kuhwan and Kang, Byungyup and Chung, Hyoju},
booktitle = {European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases},
year = {2016},
pages = {509-523},
doi = {10.1007/978-3-319-46128-1_32},
url = {https://mlanthology.org/ecmlpkdd/2016/kim2016ecmlpkdd-online/}
}