Jul 2026· Journal of Intelligence and Information Systems· 0 citations· 28 references
Abstract
As Large Language Models (LLMs) become foundational to next-generation Intelligent Information Systems, the bridge between natural language interfaces and structured database systems remains a critical bottleneck. While Text-to-SQL generation enables cooperative support for complex query formulation, ensuring the reliability of these generated queries at inference time is a central challenge. Conventional methods rely on coarse execution-based signals, which may limit their ability to capture the nuanced semantic alignment required for high-stakes database environments. In this work, we propose the use of Outcome Reward Models (ORMs) as a fine-grained, probabilistic feedback mechanism for test-time verification in Text-to-SQL tasks. We introduce GradeSQL, a framework for training task-specific ORMs that assign scalar utility scores to candidate SQL queries based on their semantic correctness and alignment with database schema. Our approach is evaluated on the BIRD and Spider benchmarks across multiple open-source LLM families. Experimental results demonstrate that ORM-based verification consistently outperforms traditional execution-based heuristics.
It is demonstrated that ORM-based verification provides a simple, effective, and scalable alternative to heuristic test-time selection strategies for Text-to-SQL, and that ORMs scale effectively with larger candidate sets and yield stronger improvements on complex queries.
M. Tritto, G. Farano, Dario Di Palma et al.· arXiv.org· 1 citation· ⚡1
SQuaD-SQL (Small-Qualified and Distilled for SQL), a novel approach that empowers small language models to approach the performance of LLMs on the Text-to-SQL task while significantly improving efficiency through knowledge distillation and synthetic data generation, is introduced.
Wangyu Wu, Xiaojian Lin, Rong Fu et al.· 0 citations
Learning Management Systems accumulate heterogeneous data comprising unstructured course documents and structured relational databases, yet most existing tools address only one data type, limiting accessibility for non-technical users. This paper proposes a dual-pipeline large language model assistant that integrates a Retrieval-Augmented Generation pipeline for document queries with a Natural Language to SQL pipeline for database interaction. The RAG pipeline deploys Phi 3.5 and nomic-embed-text entirely on local hardware to ensure institutional data privacy, while Google Gemini generates SQL from schema metadata only. Semantic chunking based on embeddingsimilarity thresholds replaces fixed character splitting, and a lightweight generate-validate-execute-correct mechanism handles SQL failures. Evaluation using the RAGAS framework shows that semantic chunking improves answer relevancy by +0.1345 over fixed splitting. Testing on the BIRD-MINIDEV benchmark yields a $\mathbf{1 0. 5 3} \boldsymbol{\%}$ relative improvement in execution accuracy with 100% valid SQL generation. The proposed system offers a practical, privacy-compliant solution suitable for institutional deployment without large-scale infrastructure.
Shreyas S Acharya, Govind Sreekar Shenoy, Sudhir Shenai et al.· 2026 5th International Confe...· 0 citations
Results show that SQL verification can be performed with a lightweight learned model while retaining feature-level evidence for inspecting and diagnosing its predictions, and feature attribution shows that the model relies on both semantic grounding and deterministic SQL-structure signals.
N. Shukla, Debasmita Panda, Srutanik Bhaduri et al.· 0 citations
This work presents a training-free framework that formulates SQL correction as a plan-guided, tree-structured debugging process that mitigates error accumulation during iterative refinement and demonstrates the effectiveness and stability of the approach in real-world deployments.
Xinmei Huang, Jie Song, Peng Li et al.· 0 citations
This work proposes MARS, a scalable knowledge graph question answering (KGQA) approach that requires no model fine-tuning, and performs a structured retrieval procedure that links question entities to the KG and iteratively retrieves relevant next-hop information.
Nikit Srivastava, Daniel Vollmers, René Speck et al.· 0 citations