The GELATO Dataset for Legislative NER
The GELATO dataset introduces a two-level named entity recognition ontology for U.S. congressional legislation from the 118th Congress, consisting of six top-level classes and thirty subclasses. Researchers fine-tuned transformer models like RoBERTa for first-level prediction, followed by using LLMs to complete the second level. The results show RoBERTa's superior performance over BERT for this task.