LogMap

LogMap's core innovation is its ability to detect and repair logical inconsistencies in real-time during the matching process. When candidate mappings cause unsatisfiable classes in the merged ontology, LogMap uses a Dowling-Gallier algorithm to compute the minimal set of conflicting correspondences. It then systematically removes the lowest-confidence mappings from this conflict set, ensuring the final alignment is coherent and logically sound without sacrificing recall.

LogMap processes large-scale biomedical ontologies like SNOMED CT and FMA through an iterative partitioning strategy:

• Lexical Indexing: Starts with high-precision string matching to establish anchor points.
• Structural Indexing: Expands from anchors using the ontology class hierarchy.
• Propositional Horn Reasoning: Efficiently computes the deductive closure to discover implicit mappings. This divide-and-conquer approach allows LogMap to handle ontologies with hundreds of thousands of classes.

LogMap computes a confidence score for every candidate mapping by combining multiple similarity metrics:

• Lexical similarity via ISub string matching.
• Structural similarity via graph-based neighborhood comparison.
• Semantic similarity via logical class overlap. These scores are used both to rank the final alignment and to guide the repair module in deciding which conflicting mappings to discard, prioritizing high-confidence correspondences.

While LogMap is domain-independent, it is optimized for the biomedical informatics ecosystem. It consistently achieves top performance in the Ontology Alignment Evaluation Initiative (OAEI) tracks for:

• Large BioMed Track: Matching FMA, NCI, and SNOMED CT.
• Anatomy Track: Aligning the Adult Mouse Anatomy to the human NCI Thesaurus. LogMap's reasoning kernel is particularly effective at handling the rich, axiomatized definitions common in OWL 2 EL biomedical ontologies.

Unlike post-processing pipelines, LogMap integrates alignment repair directly into its matching workflow. The system provides a detailed debugging output that identifies:

• Unsatisfiable classes introduced by the mapping.
• Conflict sets of mappings responsible for each inconsistency.
• Repaired alignment with the specific mappings removed. This transparency allows ontology engineers to manually review and override automatic repair decisions, blending automation with expert curation.

LogMap is distributed as an open-source Java library with a modular design that allows users to:

• Swap lexical matchers: Replace the default ISub string metric with custom label similarity functions.
• Extend reasoning: Integrate external OWL reasoners like ELK or HermiT.
• Customize filtering: Adjust the conservativity and consistency thresholds. The system also exposes a command-line interface and a graphical user interface for interactive ontology alignment debugging.

The post-matching process of detecting and removing incoherent correspondences to restore logical satisfiability. LogMap's core innovation is its repair-by-diagnosis algorithm, which identifies minimal sets of conflicting mappings (MUS) and resolves them using a voting-based selection strategy. This ensures the merged ontology contains no unsatisfiable classes or disjointness violations, a critical requirement for biomedical ontologies like SNOMED CT and the OBO Foundry.

A logical constraint stipulating that an alignment should not introduce new subsumption relationships between named classes in the original ontologies. LogMap enforces this by checking if a mapping causes a class to gain a new superclass it didn't have before. Violations indicate the alignment is over-committing to structural similarity at the expense of domain semantics, a common failure mode in pure lexical matchers.

A quantitative metric assessing the logical consistency of an alignment by measuring the ratio of unsatisfiable classes introduced after merging. LogMap uses this as its primary optimization target, iteratively refining mappings until coherence reaches 1.0 (no violations). This contrasts with purely structural metrics like tree edit distance, which ignore the logical consequences of asserted correspondences.

The core OWL property asserting two named individuals refer to the exact same real-world entity. LogMap outputs alignments using this identity link to connect equivalent classes across ontologies. However, overuse can cause transitive explosion—if A ≡ B and B ≡ C, then A ≡ C is entailed, potentially merging distinct concepts. LogMap's repair step explicitly checks for these cascading identity violations.

Mathematical functions like edit distance or the Jaccard coefficient used as primary lexical matchers. LogMap bootstraps its alignment process with these metrics to generate an initial candidate set, computing similarity over normalized labels and synonyms. However, pure string matching ignores logical axioms, which is why LogMap subsequently applies its reasoning-based repair to filter out lexically plausible but semantically incoherent mappings.

A neural ontology matching technique encoding the structural neighborhood of entities into embedding vectors. While LogMap uses symbolic logic, GCN-based methods represent a complementary paradigm: they learn latent representations where geometric proximity indicates semantic equivalence. Hybrid systems now combine LogMap's repair-by-diagnosis with GCN-generated candidate mappings to improve recall on structurally divergent but semantically equivalent concepts.