Abstract
LogMap is a high-performance ontology matching system designed to bridge the gap between massive datasets while maintaining logical consistency. In the 2016 Ontology Alignment Evaluation Initiative (OAEI) campaign, the LogMap family demonstrated its continued status as a top-tier solution by successfully participating in every evaluation track. The system offers a rare combination of massive scalability and rigorous logic-based repair, ensuring that integrated data remains computationally usable.
Introduction
In the Semantic Web ecosystem, ontology matching is the critical process of identifying correspondences between different data schemas. While many contemporary systems rely heavily on lexical heuristics—matching entities based on string similarity—this approach often leads to "logical disasters." Such failures manifest as unsatisfiable classes (concepts that cannot logically exist) or broken hierarchical structures within the merged ontology.
LogMap addresses the challenge of creating alignments that are both scalable and logically sound. The system is built upon three core principles:
- Consistency: Mappings must not introduce logical contradictions.
- Locality: Mappings should link entities with similar "neighborhoods" in their respective graphs.
- Conservativity: Mappings should not fundamentally alter the original classification of the input ontologies.
Methodology
LogMap is not a single algorithm but a suite of techniques optimized to handle ontologies containing hundreds of thousands of classes. Its architecture comprises several sophisticated components:
Lexical and Semantic Indexation
LogMap utilizes an inverted index, similar to search engine technology, to rapidly identify an initial set of candidate mappings. To handle taxonomic queries efficiently, it employs a semantic indexation strategy using an interval labeling schema. This allows the system to answer complex questions (e.g., "is concept A a sub-class of B?") as simple index lookups rather than resource-intensive reasoning tasks.
Logic-based Module Extraction and Reasoning
To maintain scalability, LogMap avoids reasoning over massive ontologies in their entirety. Instead, it extracts smaller, relevant "modules" to make unsatisfiability detection computationally feasible. Candidate mappings and these modules are encoded into a Horn propositional representation. The system then applies the Dowling-Gallier algorithm to detect logical conflicts with high efficiency.
Greedy Local Repair
When a logical conflict is identified, LogMap performs "on-the-fly" repair. This greedy approach resolves inconsistencies before finalizing the alignment, ensuring the output is logically coherent. This repair functionality has also been decoupled into a standalone service, LogMap-Repair, which can be utilized by other matching systems to improve their output quality.
System Variants
The LogMap family includes specialized versions for different use cases:
- LogMapLt: A lightweight variant focusing on rapid string matching.
- LogMapBio: A version tailored for biomedical domains that leverages external knowledge from BioPortal to enhance mapping discovery.
Evaluation: OAEI 2016
The system was rigorously evaluated during the OAEI 2016 campaign across a variety of tracks, including:
- Anatomy: Matching human and mouse anatomy ontologies.
- LargeBio: Large-scale biomedical ontologies including FMA, SNOMED CT, and NCI.
- Conference: A collection of schemas describing academic conference organization.
- Instance Matching: Identifying identical real-world objects across disparate datasets.
Key Results
LogMap's performance in the 2016 campaign highlighted several strengths:
- Versatility: It remains one of the few systems capable of participating in every OAEI track, regardless of scale or domain.
- Coherence: LogMap consistently produced "almost coherent" mappings, significantly outperforming systems that ignore logical constraints.
- Efficiency: The system demonstrated the ability to match ontologies with hundreds of thousands of classes within reasonable timeframes, a significant hurdle for traditional logic-based reasoners.
Discussion and Future Work
Despite its success, LogMap faces limitations regarding lexical dependence. Because the initial candidate generation relies on an inverted index of words, the system struggles with ontologies that are lexically disparate or those that use opaque identifiers (IDs) instead of human-readable labels.
Future development aims to improve multilingual support and deepen integration with external resources like BioPortal. Additionally, the authors are exploring the application of LogMap in high-stakes industrial environments, such as the oil and gas industry, through initiatives like the FP7 Optique project. By treating logic as a first-class citizen, LogMap continues to prove that logical rigor and massive scalability are not mutually exclusive in the field of ontology matching.
Summary updated