[Novel] KeySearchWiki: An Automatically Generated Dataset for Keyword Search over Wikidata

Leila Feddoul, Frank Löffler, Sirko Schindler

Abstract: Keyword search is an intuitive method to access knowledge graphs without requiring technical expertise or knowledge of the underlying data schema. In this context, various methods for keyword search over knowledge graphs have been developed. However, only few evaluation datasets have been created, mostly based on a time-consuming manual generation. We present KeySearchWiki, an automatically generated dataset for keyword search over Wikidata, containing over 16 thousand queries and their relevant results. It is based on Wikidata and Wikipedia set categories which are refined and combined to derive more complex queries. We explain the dataset generation workflow, highlight some dataset characteristics, present experiments using baseline retrieval methods, and evaluate the accuracy of relevant results.

Submission Number: 4