CHORUS: Foundation Models for Unified Data Discovery and Exploration
Moe Kayali, Anton Lykov, Ilias Fountalis, Nikolaos Vasiloglou, Dan Olteanu, Dan Suciu
TL;DR
This paper demonstrates that foundation models can effectively unify data discovery tasks in data lakes and warehouses. By introducing Chorus, a prompting-based framework with post-processing and anchoring, the authors tackle table-class detection, column-type annotation, and join-column prediction, achieving superior results over task-specific baselines and often rivaling human experts. The approach emphasizes zero-/few-shot prompting, cross-task information flow, and robust risk mitigations, including domain-specific anchoring. Empirical results across multiple benchmarks, contamination checks, and ablations establish the method's robustness, scalability, and potential to transform data discovery workflows in practice.
Abstract
We apply foundation models to data discovery and exploration tasks. Foundation models include large language models (LLMs) that show promising performance on a range of diverse tasks unrelated to their training. We show that these models are highly applicable to the data discovery and data exploration domain. When carefully used, they have superior capability on three representative tasks: table-class detection, column-type annotation and join-column prediction. On all three tasks, we show that a foundation-model-based approach outperforms the task-specific models and so the state of the art. Further, our approach often surpasses human-expert task performance. We investigate the fundamental characteristics of this approach including generalizability to several foundation models and the impact of non-determinism on the outputs. All in all, this suggests a future direction in which disparate data management tasks can be unified under foundation models.