Evaluation of Property Recommenders for Wikidata
Wikidata is a central storage for structured data of various projects of the Wikimedia Foundation like Wikipedia. The structured data is stored in form of subject-property-object triples, where each item of the triple can be freely chosen. Because of this freedom to choose each of these items, it is very likely that users enter synonyms and therefore, the system becomes heterogeneous and proliferated. With the Snoopy Concept, a concept was introduced, that fights against proliferation of the system and additionally pushes the user to enter more information to extend the system by exploiting the users valuable knowledge. The Snoopy Concept is designed to do this by giving various recommendations, e.g., while a user types in a new property, already existing properties are recommended to avoid synonyms.
This work focuses on the recommendation of properties and evaluates how good they perform on the dataset of Wikidata. The property recommender that is implemented in Wikidata is evaluated and compared with other recommenders that are taken into account in the SnoopyDB project, which is an implementation of the Snoopy Concept. Additionally, another approach, that can be used for the auto-amendment of triples, is presented and evaluated how good the recommendation of properties work.