MDB-Link: Hierarchical Schema Linking for Multi-Database Text-to-SQL

2026-08-10

AI Summary

This work addresses the challenge in multi-database Text-to-SQL tasks, where accurate SQL generation requires first identifying the target database. To this end, the authors propose a hierarchical schema linking framework that introduces, for the first time, a hierarchical schema reduction mechanism. This approach retrieves relevant columns via a global column index, aggregates evidence to filter candidate databases, and leverages a budget-aware large language model (e.g., Qwen2.5-14B) to perform database re-ranking, table selection, and column alignment, thereby constructing a compact, SQL-relevant schema subset. The method achieves Exact Match scores of 51.41, 9.17, and 38.01 on MMQA, Spider2-Snow, and BIRD-dev, respectively, significantly outperforming existing baselines while also improving inference efficiency.

Abstract

Traditional Text-to-SQL research and benchmarks assume a known target database, overlooking settings in which a query must be routed within a large, heterogeneous database collection. We therefore study schema linking in a multi-database setting, where the system must first locate the target database and then construct a compact, SQL-relevant schema for generation. We propose MDB-Link, a hierarchical schema-linking framework that retrieves question-relevant columns from a global index, aggregates retrieval evidence to shortlist databases, and uses a budget-aware large language model (LLM) for database reranking, table selection, and column grounding. With Qwen2.5-14B, MDB-Link outperforms LinkAlign on MMQA, Spider2-Snow, and BIRD-dev in database localization and column selection while producing schema subsets close in size to the gold schemas. Exact match improves from 16.88 to 51.41 on MMQA, 2.50 to 9.17 on Spider2-Snow, and 12.52 to 38.01 on BIRD-dev. MDB-Link also runs faster than LinkAlign and AutoLink, demonstrating the effectiveness of hierarchical schema reduction for downstream SQL generation.

Problem

• multi-database
• schema linking
• Text-to-SQL
• database localization
• heterogeneous databases

Innovation

• hierarchical schema linking
• multi-database Text-to-SQL
• schema reduction
• budget-aware LLM
• database localization