Achieving Precise Text-To-Cypher Via Grounded Knowledge Graph Data Generation

The paper introduces a synthetic data generation method aimed at enhancing small LLMs for Text-To-Cypher (Text2Cypher) tasks, allowing for improved conversational interfaces with Property Graphs. Experiments on major Text2Cypher benchmarks show that this approach significantly boosts the performance of small LLMs, enabling them to rival larger proprietary models. This advancement is crucial for practitioners focused on local model deployment, as it supports data sovereignty while maintaining accuracy without extensive annotation efforts.