Latent Dirichlet Allocation (LDA) applied to 2,000 PubMed abstracts using collapsed Gibbs sampling to discover 20 latent biomedical topics.
output quality is evaluated using topic coherence metrics (c_v & c_npmi), indicating interpretable biomedical topics.
See the full implementation and results in LDA.ipynb.
Collapsed Gibbs sampling implementation borrowed & adapted from CS179 course material by Prof. Alexander Ihler.