Lory-J: Location-Aware Join Discovery from Data Lakes

Abstract

Data lakes enable flexible, ingestion-first data management, but their lack of global schemas makes it difficult for users to discover datasets across heterogeneous tables. While prior systems focus on relational join discovery, they largely ignore geo-spatial semantics, despite geographic attributes being pervasive in tabular data. As a result, users cannot express or discover joins based on fundamental spatial relationships, e.g., proximity, containment, or intersection, to discover datasets that are co-located in space. We demonstrate Lory-J, a system for location-aware join discovery in tabular data lakes. Lory-J treats spatial attributes as first-class signals and enables the discovery of hybrid join paths that combine equi- and semantic-joins with spatial joins. Through an interactive, map-driven interface, users specify spatial constraints to discover a dataset. The demo showcases how Lory-J supports end-to-end dataset construction by discovering, visualizing, and materializing join paths driven by both relational and spatial intent—without requiring prior knowledge of table schemas or join keys.