Source-Grounded Data Generation for Text-to-JSON Learning

2026-06-18

AI Summary

Traditional industries rely on unstructured long-form documents to store high-value information, yet they lack high-quality, scalable training data for text-to-JSON structuring. This work proposes STAGE, a method that achieves the first end-to-end synthetic data generation pipeline grounded in spreadsheet source data: leveraging large language models to produce semantically coherent narrative reports alongside their corresponding JSON representations, and employing value-level verification against the original source data to ensure factual fidelity. The STAGE-Eval dataset constructed via this approach substantially enhances model performance—on Qwen3-4B, exact match accuracy improves from 31.37% to 74.27%, and value-level accuracy rises from 45.46% to 90.69%.

Abstract

From financial filings to clinical records, legacy industries rely heavily on long, unstructured documents to store high-value information. Reliably extracting this information into structured, machine-readable representations is a key prerequisite to making the contents accessible to automated systems. JSON is a natural target for such structured extraction, yet constructing reliable and scalable text-to-JSON training data remains challenging. To address this gap, we propose STAGE (Spreadsheet-grounded Text-to-JSON Artifact GEneration), a source-grounded data generation pipeline that constructs reports and JSON schema by using LLMs for scalable synthesis while validating ground-truth values against the underlying spreadsheet. Evaluations on STAGE-Eval, our source-grounded benchmark with an 851-example test set, show that STAGE produces stronger training data than existing approaches. This improves Qwen3-4B exact match from 31.37% to 74.27% and value accuracy from 45.46% to 90.69%.

Problem

• text-to-JSON
• structured extraction
• unstructured documents
• training data generation
• JSON schema

Innovation

• source-grounded generation
• text-to-JSON
• structured data extraction
• LLM-based synthesis
• spreadsheet validation