Keywords: federated queries, SPARQL, Wikidata Wikidata is the linked database of Wikipedia and other sister projects from the Wikimedia foundation. Similar to Wikipedia, Wikidata is not limited to a select set of knowledge domains, in principle it can capture all knowledge. However, external resources can contain much more granularity on a given topic. Research in the life sciences for example can generate vast amounts of data (e.g. highthroughput screening), so much that storing all in Wikidata will not be possible. More e ective would be to consider Wikidata the central hub between more detailed life-science hubs on the semantic web. We present how Wikidata can be linked to and used to link other semantic sources. We compare four federated SPARQL query patterns to create the links. To link Wikidata to the semantic web, we have followed four routes: (1) Using the FILTER operator in SPARQL; (2) Using IRIs which are compiled on the y based on existing -value based- shared identi ers, using SPARQL's BIND operator; (3) Storing Wikidata-item identi ers in external data sources as mappings to local identi ers; and (4) Storing remote IRI's as a Wikibase property value of type URL.
queries, Wikidata and an external source can be linked through a SPARQL query. Since both identi ers will be stored as literals, linking through IRIs is not an option. Using lters in SPARQL is then an option. The FILTER operator from SPARQL limits the results of a graph pattern based on boolean or regular expressions. This only works if the results of both graph patterns are not substantial in size or complexity such that the SPARQL endpoint can handle processing both returns. We demonstrate this using graph patterns from Wikidata and WikiPathways.
Using the BIND operator in SPARQL queries When, however, the external source contains IRIs for their identi able concepts, but the equivalent Wikidata property is still available only as a value, a link can be made using the BIND operator, where an IRI is composed based on the property value stored in Wikidata. We demonstrate this using the BIND operator by a federated query which connects Wikidata with Uniprot, allowing merges of protein annotations from both Wikidata and Uniprot.
External resources, like WikiPathways and DisGeNET, actively store mappings between their concepts and the equivalent items in Wikidata. When resources maintain such mappings, a federated query where the remote query pattern and the Wikidata query pattern share the same variables is su cient to merge results.
We have proposed a Wikidata property (P2888) that allows to create mappings between Wikidata and external resources on the semantic web as Wikidata statements. This property has been accepted and we are currently populating items with statements expressing similarity between Wikidata items and others concept descriptions on the semantic web. This Wikidata property is based on skos:exactMatch.
Linking Wikidata to other hubs on the semantic web opens up new routes of validation between di erent resources and toward better integration. We have demonstrated four potential paths to link content from Wikidata with other resources on the semantic web, using federated SPARQL queries.