Benchmark Datasets for Entity Linking from Tabular Data

This archive provides a benchmark suite for evaluating entity linking algorithms on structured tabular data. It includes five datasets: HTR1 and HTR2: From the SemTab Table-to-KG Challenge, commonly used in academic evaluations of table-to-KG alignment systems. Movies and Companies: Constructed using SPARQL queries on Wikidata and enriched using LamAPI, which retrieves object and literal relationships to create well-formed relational tables. SN (Spend Network): A real-world procurement dataset from the enRichMyData project. It has been manually annotated with Wikidata QIDs and includes NIL cases for mentions with no known Wikidata match. ⸻ Each dataset contains: A tabular input file (CSV) A ground truth file mapping mentions to Wikidata entities A file mapping each mention to a list of candidate Wikidata QIDs, including the expected correct entity ⸻ 📦 Archive structure and details are provided in the included README.md. 📝 Licensing: • HTR1 and HTR2: CC BY 4.0 • Movies and Companies: Derived from Wikidata (public domain, CC0 1.0) • SN: Licensed under CC BY 4.0 ⸻ If you use these datasets in your work, please cite the respective data sources: SemTab, Wikidata, and enRichMyData.