Large Language Models for Data Extraction in Slot-Filling Tasks
摘要
Large language models (LLMs) have turned out recently to be a powerful tool for solving natural language processing and understanding tasks. In this paper, we investigate the usage of three open-source large language models in a slot-filling task, which is a crucial task in chatbot development. Apart from testing the method on an in-house created dataset, we checked the methodology on two main benchmarks in this field. The obtained results for models with 7B parameters are comparable with those achieved by closed-source chatGPT family models, which are more than 20 times bigger.