Wikidata data-quality

Overview

• example query
• contributor effort
• documentation by example
• documentation by definition
• ShEx features

# List of present-day countries and capital(s)
SELECT DISTINCT ?country ?countryLabel ?capital ?capitalLabel
WHERE
{
    ?country wdt:P31 wd:Q3624078 .
    # not a former country
    FILTER NOT EXISTS {?country wdt:P31 wd:Q3024240}
    # and no an ancient civilisation (needed to exclude ancient Egypt)
    FILTER NOT EXISTS {?country wdt:P31 wd:Q28171280}
    OPTIONAL { ?country wdt:P36 ?capital } .
    SERVICE wikibase:label { bd:serviceParam wikibase:language "en" }
}
ORDER BY ?countryLabel

contributor sisyphean task

Given some primary source, how to model in WD:

• guess from available properties (~10K)
• guess from survey of existing data
  • what you produce depends on what you happen to survey
  • reinforces arbitrary modeling choices
• launch grueling discussion for any needed novel properties

documentation by example

• contributor is modeling each "column" in isolation
• no guidance for domain-specific links between resource types
• rigorous survey can take hours or days of expert time

no central place to establish/review community practice

documentation by definition

• provides guidance for constellations of entity structures with:
  • properties
  • datatypes
  • cardinalities
• first-class WD entities with edit logs and discussion
• communities with different use cases can have their own schemas
  • small number of well-defined competing schemas
  • NOT fuzzy epicenters of modeling habits with unclear advantages

IMPORT <E91>

start=@<#FLOSSemulator>
<#FLOSSemulator>
  EXTENDS @<E91#emulator>
  EXTRA p:P31  {
    # copyright license
    p:P275 @<license> +;
    # source code repository;
    p:P1324 { ps:P1324 IRI } ?  ;
    # Framalibre ID;
    wdt:P4107 xsd:string ? ;
    # Pro-Linux.de DBApp ID; 
    wdt:P6665 xsd:string ? ;
    # SWH Release ID; 
    wdt:P6138 xsd:string ? ;
    …
}

<#license> {
     p:P31 [wd:Q3943414]
}