SALT-KG: A Benchmark for Semantics-Aware Learning on Enterprise Tables

Isaiah Onando Mulang', Felix Sasaki, Tassilo Klein, Jonas Kolk, Nikolay Grechanov, Johannes Hoffart

Submission Type: Short paper (4 pages)

Keywords: Knowledge Graph, Foundation Models, Table Representation Learning, Enterprise Data, Business Knowledge, Semantically Linked Tables, Foundation Models for Semantically Linked Tables, FMSLT

TL;DR: A semantically linked tables benchmark dataset for predictive learning and foundation models for tabular data. Semantic Links from a metadata knowledge graph (KG) are aligned to the underlying relational schema dataset.

Abstract:

Building upon the SALT benchmark for relational prediction, we introduce SALT-KG, a benchmark for semantics-aware learning on enterprise tables. SALT-KG extends SALT by linking its multi-table transactional data with a structured Operational Business Knowledge represented in a Metadata Knowledge Graph (OBKG) that captures field-level descriptions, relational dependencies, and object-type hierarchies. This extension enables controlled evaluation of models that jointly reason over tabular evidence and contextual semantics—an increasingly critical capability for foundation models on structured data. Empirical analysis reveals that while metadata-derived features yield modest improvements in classical prediction metrics, these metadata features consistently highlight gaps in models’ ability to leverage semantics in relational context.
By reframing tabular prediction as semantics-conditioned reasoning, SALT-KG establishes a benchmark to advance tabular FMs grounded in declarative knowledge, providing the first empirical step toward semantically linked tables in structured data at enterprise scale

Submission Number: 40