Language Model Comparison for Text Classification

The Interagency Implementation and Advanced Concepts Team (IMPACT) evaluated two language models (GPT-3 and BERT-E) to classify SNWG survey responses into thematic areas so that the responses can be automatically forwarded to the right Subject Matter Experts. The domain specific model (BERT-E) significantly outperforms the general GPT-3 language model and thereby demonstrating the need to build domain specific models for science.

Impact Logo 2
Scroll to Top