The Fundamentals of
Semantic Versioned Querying

Versioning Example

Version A:
    ex:Bob a ex:Animal.
    ex:Bob foaf:name "Bob".

Version B:
    ex:Bob a ex:Cat.
    ex:Bob foaf:name "Bob".

Language:
    ex:Cat rdf:subClassOf ex:Animal.

- ex:Bob a ex:Animal.
+ ex:Bob a ex:Cat.

+ ex:Bob a ex:Cat.

RDF Versioning

Semantic query atoms

Semantic extension of the regular query atoms by adding a versioned language

• Semantic Version Materialization (S-VM):
  
  Query over the semantic closure of a single version for a language version.
• Semantic Delta Materialization (S-DM):
  
  Query differences between the semantic closure of 2 versions w.r.t. language versions.
• Semantic Version Query (S-VQ):
  
  Query the semantic closure of all versions w.r.t. a language version,
  results are annotated with the version.

OSTRICH: a versioned triple store

• Compressed storage of multiple versions
• Supports stream-based structural VM, DM and VQ triple pattern queries

Implementing the semantic query atoms
on top of OSTRICH

• Separated storage of data and language versions
• Backwards rule-based reasoner accepts rules in Notation 3 syntax
• Iterative query-time reasoning: 1 semantic query => multiple structural queries

Evaluation using the BEAR benchmark

• Goal: Compare pre-inferred dataset with inference at query-time.
• We created a derived version of the BEAR-B-daily dataset.
• • Removal of rdf:type relations inferrable through rdfs:subClassOf.
  • Original: 48,914 triples in 88 versions
  • Derived: 31,761 triples in 88 versions (35,07% reduction)

~8 structural queries needed
per semantic query

Conclusion: Combining reasoning and versioned querying is beneficial

• + Not storing inferrable knowledge reduces storage requirements

• + Flexible range of knowledge by applying different reasoning rules

• - Inference increases slows down querying

  Our wrapper-based prototype is sub-optimal, room for optimizations
• • Inspiration from related domains: OBDA, RDF Stream Reasoning
  • Native implementation