Quality control tools and procedures

• Perception: Anybody can edit anything - so Wikidata is no reliable source of knowledge
• Seen as a threat for information systems based on Wikidata
  • particularly by some large Wikipedias (e.g., the English one)
• Basic policy to address this: Statements should be referenced

QA support for editors

• Contraint definition for properties
  • raise warnings during data input, when, e.g.
    • a format definition (ISBN, DOI etc.) is violated
    • a supposedly unique identifier is added to more than one item
  • generated lists of constraint violations (e.g. ZDB ID format)
• Constraints can be very helpful, but do not cover complex cases

More QA support for editors

• Additional reports can be created via SPARQL queries
• Shape Expressions (ShEx) allow to define complex constraints and conformance checks
  • ShEx Primer
  • How to get started with Shape Expressions in Wikidata?

Revision control and patroling

• Versioned edits and version control
• Manual and tool supported vandalism prevention
• Watchlists
• Automated flagging of suspect edits (e.g., “new editor deleteing statements”)
• Patroling
• Technically very easy to revert edits
• Semi-protection or protection of oftenly-vandalized items

Automated tools for vandalism detection

• Fighting to keep up with rate of human edits in Wikidata (multiple per second)
• … requires reducing the manual workload, e.g. via
  • Wikidata Abuse Filter
  • Objective Revision Evaluation Service (ORES)
• and other rule-based and machine-learning tools

Ongoing research

• Heindorf et al.: Vandalism Detection in Wikidata
• Sarabadani et al.: Building automated vandalism detection tools for Wikidata