Enterprise data analytics requires linking analytical requests with knowledge scattered across databases, business documents, and existing example tasks. Without an explicit representation of this knowledge, an LLM agent may produce erroneous results and struggle to incorporate feedback reliably. AgentOnto addresses this by organizing heterogeneous enterprise knowledge into an ontology that directly guides the analysis workflow and integrates confirmed feedback.
During the Ontology Preparation phase, the system generates its own analysis tasks to iteratively refine the ontology and improve concept mappings. Evaluation on two public datasets shows that AgentOnto substantially outperforms representative baselines in both accuracy and robustness. Module‑level ablations confirm the contribution of ontology construction, analysis execution, and ontology updating. A real‑world deployment further demonstrates practical value in enterprise analytics.
These findings indicate that explicitly constructing, leveraging, and continuously updating enterprise knowledge supports more reliable and adaptable data analysis.
Review