A Pattern-based Ontology Matching Approach for Detecting Complex Correspondences

Dominique Ritze, Christian Meilicke, Ondrej Sváb-Zamazal, Heiner StuckenschmidtOM 2009 · OpenReview

Introduction

Ontology matching is a fundamental task in data integration, aiming to identify correspondences between different conceptual schemas. While existing tools excel at finding simple 1-to-1 mappings (e.g., realizing that Author in one system is the same as Writer in another), they often fail to capture "complex" correspondences where a concept in one ontology relates to a combination of terms in another.

This semantic heterogeneity—where different engineers model the same domain using varying levels of granularity or different structural patterns—remains a significant bottleneck for automated systems. For example, one ontology might define a specific subclass, while another uses a general class combined with a boolean property to represent the same information.

Figure 1: Examples of two ontologies ($O_1$ and $O_2$) illustrating different modeling choices for the same domain, such as the use of subclasses versus properties.

Methodology: Pattern-Based Detection

To address this, we propose a structured, pattern-based approach that identifies four recurring types of complex correspondences:

  1. Class by Attribute Type (CAT): A specific class (e.g., AcceptedPaper) is mapped to a general class restricted by a related concept (e.g., Paper with a hasDecision of Acceptance).
  2. Class by Inverse Attribute Type (CAT⁻¹): A class is defined by the "reverse" of a property (e.g., a Researcher is a Person who is the object of a researchedBy property).
  3. Class by Attribute Value (CAV): A class is mapped to a property with a specific value, often a boolean flag (e.g., SubmittedPaper ≡ Paper where submission is true).
  4. Property Chain (PC): A single property is represented as a sequence of two or more properties (e.g., authorName ≡ hasAuthor followed by name).

Our detection framework employs a hybrid logic combining syntactical and structural heuristics. A key innovation is the use of "head noun" extraction. For instance, in the term AcceptedPaper, the algorithm identifies Paper as the head noun and Accepted as the modifier.

Figure 4: Detection logic for a CAT correspondence, showing how the system matches the head noun 'Paper' and the modifier 'Accepted' to related structures in the target ontology.

As shown in Figure 4, the system verifies the mapping by checking if the head noun matches a class in the target ontology and if the modifier has a semantically similar counterpart within the range of a relevant property. This structural check ensures that the proposed mapping makes sense within the existing class/subclass tree.

Evaluation and Results

The approach was evaluated using real-world and bibliographic datasets from the Ontology Alignment Evaluation Initiative (OAEI), including the Conference and Benchmark tracks.

  • Benchmark Performance: In bibliographic datasets, the algorithm demonstrated high stability in detecting Property Chains, identifying 17-18 correct mappings with high precision.
  • Conference Domain: In more diverse conference ontologies, CAT patterns were more prevalent, highlighting that complex modeling styles are highly domain-dependent.
  • Impact: The method increased the total number of identified correspondences by approximately 4% to 10% compared to simple matching alone. While lower similarity thresholds (e.g., 0.6) increased recall, a threshold of 0.8 provided the most balanced precision and recall.

Conclusion

By moving beyond simple string matching to a pattern-based structural analysis, this approach successfully uncovers sophisticated relationships in heterogeneous data. While the system significantly enriches alignments, the authors note that human verification remains important due to the mathematically "explosive" search space of potential property combinations. Future work will focus on expanding the system to handle non-boolean attribute values and reducing the dependency on the accuracy of initial simple alignments.

Summary updated

BibTeX
@inproceedings{ritze2009pattern,
  author = {Dominique Ritze and Christian Meilicke and Ondrej Sváb-Zamazal and Heiner Stuckenschmidt},
  title = {{A Pattern-based Ontology Matching Approach for Detecting Complex Correspondences}},
  booktitle = {OM},
  year = {2009},
  url = {https://openreview.net/forum?id=6qu5DmpHpZw},
}
Figure