Khai thác mô hình ngôn ngữ lớn để chuyển đổi ngôn ngữ tự nhiên thành truy vấn Cypher một cách hiệu quả

Abstract

This paper investigates the application of large language models, specifically GPT, in the task of transforming natural language into Cypher queries (Text-to-Cypher). This is a critical component in enhancing chatbot systems based on graph databases. We analyze prominent methods, including zero-shot, few-shot, and fine-tuning approaches, and propose an improved model for the few-shot method. Finally, we evaluate their effectiveness in transforming natural language inputs into Cypher queries with high accuracy and efficiency. By analyzing performance across various scenarios, the paper highlights the trade-offs between generalization, accuracy, and resource requirements. The findings emphasize the growing importance of Text-to-Cypher tasks in advancing AI-driven conversational technologies.