Comparison of LLM Models and Strategies for Structured Query Construction from Natural Language Queries

The advent of large language models (LLM) has sharply increased the power of machine translation because of their remarkable success in many text processing tasks. This feature makes LLMs attractive options for specific tasks, such as translating natural language queries into keyword-based queries. Within this research, we compare the different techniques applied in LLMs for open and closed-source model options to facilitate decision-making for researchers and practitioners. Our findings present that while fine-tuning is a highly recommended option, prompting techniques allow us to achieve similar performance with commercial tools; however, before any model selection, it is essential to consider the amount of data, computational resources, and data privacy.