ctm_topics

Introduction

The ctm_topics container adds contextualized topic modeling to the framework. The model is taken from https://github.com/MilaNLProc/contextualized-topic-models. We have (not yet) made any adjustments to the model. It is copied “as is”. We only provide a (smart) wrapper to access the models. “Smart” means:

• you can define runtime variables in settings.txt, e.g. 
  • number of chunks the model is continuously trained
  • define ctm model, either German or English model
  • set number of epoch the model is trained
  • set number of prediction samples

Use this model if (almost) no apriori knowledge of the topics to be discussed is known. If (generic) topics are known beforehand, you might want to use the module a2t.

Technical Framework

The wrapper is written in R and uses reticulate package to access contextualized topic model Python scripts.

Main Contributor

Valentin Gold