Hierarchical Latent Dirichlet Allocation (hLDA) addresses the problem of learning topic hierarchies from data. The model relies on a non-parametric prior called the nested Chinese restaurant process, which allows for arbitrarily large branching factors and readily accommodates growing data collections. The hLDA model combines this prior with a likelihood that is based on a hierarchical variant of latent Dirichlet allocation.
Hierarchical Topic Models and the Nested Chinese Restaurant Process
The Nested Chinese Restaurant Process and Bayesian Nonparametric Inference of Topic Hierarchies
- hlda/sampler.py is the Gibbs sampler for hLDA inference, based on the implementation from Mallet having a fixed depth on the nCRP tree.
- Simply use
pip install hldato install the package. - An example notebook that infers the hierarchical topics on the BBC Insight corpus can be found in notebooks/bbc_test.ipynb.
From a basic anaconda environment (2020.02), a number of additional packages need to be installed for the notebook to work out of the box:
conda update -n base -c defaults condaconda install jupyter matplotlib nltk wordcloud
Black:
conda install black- https://neuralcoder.science/Black-Jupyter/
Tomotopy:
conda install py-cpuinfopip install tomotopy
Also make sure the necessary NLTK packages are downloaded and accessible.