AI research
A research paper titled "UniDataAgent: An Ontology--Grounded Agent for Enterprise Question--to--Report Automation" (arXiv:2609.27257), published on September 23, 2026, by researchers from China Unicom Software Research Institute and Nankai University, presents an ontology--grounded framework designed to overcome the semantic failures of traditional retrieval--augmented generation (RAG) in corporate analytics. Rather than relying on unstructured text embeddings or direct question--to--SQL translation, UniDataAgent decouples semantic acquisition from online analytical execution. The architecture operates across two primary phases: the Ontology Acquisition and Validation (OAV) stage, which synthesizes versioned domain ontologies from metadata and business rules in hours rather than weeks, and the Question--to--Report Execution (QRE) stage, which dynamically retrieves formal semantic contracts to guide LLM tool use and generate verified, evidence--linked reports. In real--world enterprise deployments across 27 operational tables and thousands of metric variants, the ontology--grounded system attained 95.0% strict reasoning accuracy, substantially outperforming traditional document RAG baselines at 72.5%.