Abstract
LogMap is a highly scalable, logic-based ontology matching system designed to link massive datasets while automatically repairing logical inconsistencies. In the OAEI 2012 campaign, LogMap and its variant LogMapLt demonstrated that by using lightweight logic approximations and smart indexing, it can perform complex alignments in minutes—tasks that would take standard reasoners weeks to process.
Introduction
Ontology matching—finding correspondences between entities in different knowledge frameworks—is a cornerstone of the Semantic Web. However, two major "walls" hinder large-scale matching: the quadratic explosion of potential pairs (e.g., matching FMA and NCI involves over 4 billion pairs) and logical incoherence. Merging ontologies often creates "unsatisfiable classes" or logical contradictions that standard reasoning tools, like HermiT, are too slow to detect and repair in real-time for large-scale integrations.
Methodology
The authors present LogMap and its stripped-down sibling, LogMapLt. The core system employs several novel strategies to maintain speed without sacrificing logical integrity:
- Lexical Indexation: Uses an inverted index to quickly identify a manageable set of candidate mappings based on labels, bypassing full quadratic comparisons.
- Logic-based Module Extraction: Extracts smaller "modules" containing relevant semantic context for the entities being matched rather than processing the entire ontology.
- Propositional Horn Reasoning: Represents ontologies using Horn propositional logic. This allows the use of the Dowling-Gallier algorithm, which detects unsatisfiable classes in linear time.
- Axiom Tracking and Greedy Repair: When a contradiction is found, the system tracks the causal mappings and applies a greedy repair to resolve the conflict efficiently.
- Semantic Indexation: Uses an interval labeling schema to store hierarchies, allowing taxonomic queries to be answered via simple index lookups.
Evaluation and Key Results
The system was evaluated across diverse tracks of the Ontology Alignment Evaluation Initiative (OAEI) 2012, including Benchmark, Anatomy, Conference, and the Large BioMed track.
Large BioMed Track
This track served as the primary "stress test," involving FMA, SNOMED CT, and NCI. LogMap excelled in the SNOMED-NCI task, the largest in the track, completing the alignment in roughly 15-25 minutes. Standard reasoning to find all justifications for errors in this track would typically take over six weeks.
Instance Matching
New for 2012, this track focused on matching individual data points (ABoxes). LogMap showed promising results in the Sandbox and IIMB tasks, with F-measures ranging from 0.92 to 0.94, proving its lexical indexing techniques translate well to data-level matching.
Discussion
LogMap demonstrates that approximation is key for scalability. By using Horn logic, the system identifies the vast majority of logical errors in a fraction of the time. Furthermore, LogMapLt serves as a strong baseline, showing that high-quality lexical matching provides a significant portion of the necessary results for many tasks.
However, the system remains heavily dependent on label similarity. If ontologies lack rich labels or use random strings for IDs, the candidate mapping phase fails. Additionally, the system performed poorly on the Multifarm track due to a lack of specific multilingual translation or cross-lingual matching techniques. Future work will focus on implementing multilingual features and more sophisticated instance matching to handle complex data structures beyond simple property assertions.
Summary updated