AI research
The research paper "EvoOntology: A Self--Evolving Ontology Layer for Data Agents" (arXiv:2609.15779), published on September 14, 2026, by researchers at Renmin University of China, introduces an architectural framework bridging the semantic gap between Large Language Model agents and heterogeneous enterprise data sources. Rather than injecting static schemas into prompts or relying on unstructured exploratory queries, EvoOntology encapsulates domain ontologies into a dynamic Model Context Protocol (MCP) server structured across schema, content, and tool layers. The framework deploys an autonomous builder agent paired with a self--evolution loop that refines conceptual mappings through attribution--guided typed edits validated via backbone--conditional paired evaluation, significantly boosting factual reasoning accuracy and context efficiency across multiple LLM backbones.