During the [[ https://github.com/elixir-europe/BioHackathon-projects-2020/tree/master/projects/35 |
where we are working on subsetting wikidata, we ran into the issue of blank notes being used in the RDF of Wikidata to express unknown and no values. Unfortunately, this isn't consistent because blank notes are also used to express other things such as owl:complementOf (e.g. Q42).
These blank nodes are also problematic for anything that traverses wikidata node-by-node such as faceted browsers or ShEx validators.
It is not explicitly incorrect to have blank nodes in RDF data, but it is:
1. inconsistent with the approach that Wikidata has taken (which is to avoid blank nodes)
2. ambiguous because in RDF, blank nodes do not imply unknown values, they are simply *unidentified* nodes in the graph.
Steps to Reproduce:
* GET http://www.wikidata.org/entity/Q313093.ttl
* look for "_:" (currently _:2d22892344b969be376b57170b5e495f)
* try a SPARQL query for all properties of that node
``` SELECT ?P ?o { _:2d22892344b969be376b57170b5e495f ?p ?o }```
* Because of the semantics of SPARQL, this will try to get every triple in the database.
It is not explicitly incorrect to have blank nodes in RDF data, but it is
1. inconsistent with the approach that Wikidata has taken (which is to avoid blank nodes)
2. ambiguous because in RDF, blank nodes do not imply unknown values, they are simply *unidentified* nodes in the graph.
Remedy:
Invent a system-wide identifier for unknown values and use that Q identifier for all references to unknown value.