An Intelligent Querying System for Educational Statistics Using Pre-trained and Large Language Models

Educational statistics play a crucial role in comprehending the current state of educational development and informing decision-making processes. However, traditional data querying systems, which heavily rely on precise SQL statements, often fail to adequately support non-expert users. To address these limitations, we propose an intelligent querying system that leverages a pre-trained language model and a large language model to facilitate natural language interactions for accessing educational statistical data. Our system consists of two main components: query element parsing and SQL statement generation. During the parsing phase, a fine-tuned PLM identifies and correlates relevant tables and elements from the database schema. In the subsequent generation phase, these matched tables and elements, along with the original query, are utilized by LLMs to generate SQL statements. This methodology improves the reliability of SQL generation while mitigating the complexities associated with prompt engineering and reducing the significant costs associated with LLM fine-tuning. The proposed system has been successfully implemented within relevant business units, demonstrating its practicality and effectiveness.