/unsupervised_topic_segmentation

Primary LanguagePythonMIT LicenseMIT

Unsupervised Topic Segmentation of Meetings with BERT Embeddings

This is the code for the paper Unsupervised Topic Segmentation of Meetings with BERT Embeddings.

The code doesn't require training and uses a pretrained model from https://huggingface.co/transformers/model_doc/roberta.html See paper appendix for more information.

The entry point is eval.eval_topic_segmentation that returns the scores reported in the paper for the different methods.