Researchers propose explainable header-centric framework for semantic table interpretation and data quality assessment

A new arXiv preprint proposes an explainable, header-centric framework designed to improve the quality of tabular metadata used in knowledge graph (KG) preparation. The work targets metadata-only Semantic Table Interpretation (STI) scenarios, where cell values are missing, noisy, or unsuitable, making column headers the primary source of semantic evidence.

The framework maps column headers to 39 interpretable FinalFormat types using curated lexical resources and maintains token-level traceability through SourceKeywords. Each assigned type triggers validation rules grounded in a taxonomy of Data Quality Issues (DQIs), enabling detections such as missing data, duplicates, domain violations, wrong data type, and temporal mismatch.

The authors evaluate the framework across a diverse set of benchmarks, including UCI, Prague, Kaggle, VizNet/Sato, SOTAB, T2Dv2, and the SemTab 2024 Metadata-to-KG track, covering approximately 120,000 header columns. Results indicate broad practical coverage across noisy real-world metadata, with a parallel KG-mapping pathway supporting alignment to DBpedia and Schema.org.

The paper introduces HeadersIQ, a lightweight, unweighted data source-level quality metric derived from aggregated DQI detections. While the official GT-strict evaluation on the SemTab 2024 Metadata-to-KG track was modest, a blinded diagnostic audit attributes many prediction mismatches to benchmark granularity, aliasing, and ontology-selection effects rather than incorrect header-centric predictions.

The authors emphasize that the audit is presented as diagnostic evidence of disagreement patterns, not as a revised benchmark performance. Overall, the work presents a reusable workflow for metadata-driven semantic annotation, data source-level quality monitoring, and KG-oriented benchmark diagnosis.