Dependent Hierarchical Normalized Random Measures for Dynamic Topic Modeling
This work addresses dynamic topic modeling for text analysis applications, representing an incremental improvement with specific gains.
The authors tackled dynamic topic modeling by developing dependent hierarchical normalized random measures with power law properties, achieving superior perplexity scores on news, blogs, academic, and Twitter datasets compared to previous models.
We develop dependent hierarchical normalized random measures and apply them to dynamic topic modeling. The dependency arises via superposition, subsampling and point transition on the underlying Poisson processes of these measures. The measures used include normalised generalised Gamma processes that demonstrate power law properties, unlike Dirichlet processes used previously in dynamic topic modeling. Inference for the model includes adapting a recently developed slice sampler to directly manipulate the underlying Poisson process. Experiments performed on news, blogs, academic and Twitter collections demonstrate the technique gives superior perplexity over a number of previous models.