Wikidata:WikiProject Ontology/Issues

This is an overview of the main ontology issues found in Wikidata (the following classification is copied from File:WikidataCon 2021 - Overview of ontology issues.pdf).

Classification

• semantic drift
• structural bugs
  • "subclass of" cycles
  • mix-up of meta levels
  • redundant relations
    • redundant classification
    • redundant generalisation
  • exchanged sub-/superclasses
• upper level ontology is messy
• conceptual ambiguity
• inconsistent modeling
• overgeneralisation
• conflicting real-world models
• unclassified items

Semantic drift

• subclass of (P279) is assumed to be transitive: it holds between different levels of the class hierarchy
• Semantic drift shows when the inferences turn out to be wrong
• Individual subclass relations might be acceptable, but the combination is not
• Caused by concepts having different aspects that are merged into one:
  • e.g. mason the person vs. mason the profession

Structural bugs

"Subclass of" cycles

• Created if class A has a subclass B and B is a superclass of A
• Make it impossible to determine which items are meant to be more specific or general than others
• Amounts to declaring that the classes A and B in a hierarchy are equivalent

Mix-up of meta levels

• Occurs when, through inconsistent use of instance of (P31) vs. subclass of (P279), the same item is simultaneously a class and a metaclass, or similar.
• Brasileiro et al. (2016):
  • Z is both instance of and subclass of A
  • C has direct superclasses A and B such that B is instance of A
  • C is instance of both A and B, B is instance of A

Redundant relations

• Redundant classification
  • an item is both an instance of a class and one of its super classes.
  • If A is instance of B, which is subclass of C, then A instance of C is redundant
• Redundant generalisation
  • an item is both a subclass of a class and one of its super classes.
  • If A is subclass of B, which is subclass of C, then A subclass of C is redundant
• Locality of editing: not seeing all the consequences of one's actions
• Potentially competing needs: sometimes the “shortcut statement” may be needed

Upper level ontology is messy

• Upper ontology is hard
• The top-class entity (Q35120) has 59 direct subclasses (in 2021)
• Messy connections in the upper ontology lead to:
  • issues with automated inferencing
  • nonsensical conclusions
• People care more about local ontologies

Conceptual ambiguity

• Is caused by conceptual overloading of entities
• Makes it hard to understand what statements refer to
• Partly inherited from Wikipedia
• Partly created to integrate viewpoints
• Easier to keep overloading than to split (convenience)
• Alternative would be worse (significant increase in the number of items)

Inconsistent modeling

• Occurs when similar kinds of data is modelled in different ways
• Observable both across domains and within a single domain
  • e.g. mauve an instance of color and a subclass of one of its instances; what are colors?!
• Lack of common domain understanding?
• Several different ways to model the same data
• Very different design decisions taken for different domains

Overgeneralisation

• Instances are too high in the class tree
• Classification is too general
  • e.g. Club-Mate (Q53) is a trademark, but it would be better classified as a "food brand", which is a "brand", which is "trademark", too.

Conflicting real-world models

• Real world is a mess
• Different groups have different views on the world
• May lead to overlapping and conflicting classifications
• Qualifiers to the rescue?

Unclassified items

• Items with no classifying statements
• Not connected to existing ontology
• Often happening when new items are automatically created for new articles in Wikimedia projects

Presentations

• File:WikidataCon 2021 - Overview of ontology issues.pdf