The quality of a Knowledge Graph (KG) hinges not only on downstream graph validation but also on the quality of the tabular metadata used before integration. In metadata‑only Semantic Table Interpretation (STI), where cell values are missing, noisy, or unsuitable, column headers become the primary source of semantic evidence for traceable KG construction. This paper introduces an explainable, header‑centric framework that focuses on metadata‑only Column Type Annotation (CTA) and Data Quality Assessment (DQA).
The framework first maps each header to one of 39 interpretable FinalFormat types using curated lexical resources, preserving token‑level traceability via SourceKeywords. Based on a taxonomy of Data Quality Issues (DQIs), it then activates validation rules that detect missing data, duplicates, domain violations, wrong data types, and temporal mismatches. All detections are aggregated into HeadersIQ, a lightweight, unweighted data‑source‑level quality metric.
We evaluated the approach across heterogeneous benchmarks—including UCI, Prague, Kaggle, VizNet/Sato, SOTAB, T2Dv2, and the SemTab 2024 Metadata‑to‑KG track—covering roughly 120,000 header columns. Results demonstrate broad practical coverage on noisy real‑world metadata, while a parallel KG‑mapping pathway supports alignment with DBpedia and Schema.org.
On the SemTab 2024 Metadata‑to‑KG track, the official GT‑strict evaluation was modest. However, a blinded diagnostic audit revealed that many mismatches stem from benchmark granularity, aliasing, and ontology‑selection effects rather than fundamentally implausible header‑centric predictions. We present this audit as diagnostic evidence of disagreement patterns, not as a revised benchmark score.
Overall, the paper delivers a reusable workflow for metadata‑driven semantic annotation, data‑source‑level quality monitoring, and KG‑oriented benchmark diagnosis.
Review: By combining structured lexical mapping with explainable quality rules, the framework enables semantic inference and quality assessment even when cell values are unavailable, offering a practical front‑end for large‑scale KG construction.