Semantic Integration: A Survey of Ontology-Based Approaches

Noy

semantic integration ontology translation

@article{noy:sigmod-2004,
  title={Semantic Integration: A Survey of Ontology-Based Approaches},
  author={Noy, N.F.},
  journal={{ACM} SIGMOD Record},
  volume={33},
  number={4},
  pages={70},
  year={2004},
  publisher={{ACM}}
}

Techniques for finding, declaratively specifying, and using correspondences between ontologies

Semantic heterogeneity in structured data

All agreeing on one or even small set of ontologies is unrealistic

Ontologies may mismatch at variety of levels

• Language level: Underlying ontology languages differ in expressiveness
  • May require normalizing into the same language
• Ontology level: Discrepancies betwen concepts, terms, coverage, etc

"Many researchers agree that one of the major bottleneck(s) in semantic integration is mapping discovery."

Two major approaches

• One approach: Agree on a general upper ontology that is then extended in consistent fashion
  • Different from traditional information integration approach of developing global schema afterward
  • Use that core ontology to answer questions and semi-automatically derive mappings (PSL!)
• Other: Heuristics-based or machine learning to identify common lexical or formal structures, instances, etc
  • XML and database schema matching are similar but tend to rely less on explicit semantics
  • Ontologies generally have more constraints which may be utilized in this process

Hovy

• Natural language processing to split composite words, then comparing substrings to find similarities
• Also utilizes the natural language definitions of concepts

PROMPT

• User guides process and system assists
  • E.g., user binds two concepts, system looks at structures of sub-classes and properties to try and align more
• Also applies graph theory to concept/link graph to look for similarities

Euzenat and Volchev

• Weighted heuristic over similarities of various features defined in OWL

FCA-Merge

• Uses ontologies that have a shared set of instances (e.g., annotated documents), then applies FCA to organize those instances, and compares to the ontology structures to find similarities

IF-Map

• Information flow?

Giunchiglia & Shvaiko

• Grounding ontology in WordNet, run SAT prover over that to determine relationships such as generalization

Schema & DB mappings usually expressed as views or queries; ontology languages allow for more expressive mappings

OntoMerge applies bridging axioms in general purpose reasoner to relate clauses and properties

May have an ontology of mappings, e.g., the concept relations, and transformation functions to apply, e.g., to change values, many-to-one aggregation, etc

Of note:

• Klein---different types of ontology mismatch [16]
• Workshop on Core Ontologies in Ontology Engineering, 2004
• Y. Kalfoglou and M. Schorlemmer. Ontology mapping: the state of the art. Knowledge Engineering Review, 18(1):1--31, 2003.