Saturn: Agentic Natural Language-Driven Tabular Data Discovery
Accepted in VLDB 2027 (to appear).
Recommended citation: Zuohan Wu, Jiachuan Wang, Jiake Ge, Xujia Li, Libin Zheng, Yongqi Zhang, Shuangyin Li, Lei Chen. "Saturn: Agentic Natural Language-Driven Tabular Data Discovery." Proceedings of the VLDB Endowment (VLDB 2027), accepted; to appear.
Overview
Saturn is an expert-inspired, training-free pipeline for natural-language-driven tabular data discovery. A multi-agent framework constructs a User Preference-based Ontology (UPO) from query requirements and table evidence. Saturn reuses the ontology’s concept-specific procedures to create compact table profiles and aligns queries to the same profile space for retrieval, without task-specific alignment training.
Publication Details
- Venue: VLDB 2027
- Status: Accepted; to appear
- Accepted: October 2026
- Code: GitHub
Authors
Zuohan Wu, Jiachuan Wang, Jiake Ge, Xujia Li*, Libin Zheng, Yongqi Zhang, Shuangyin Li, Lei Chen
*Corresponding author.