Introduction
In the evolving landscape of the Semantic Web, ontologies serve as the foundational "glue" enabling heterogeneous systems to share and interpret data. However, the decentralized nature of the web often leads to the creation of distinct ontologies for identical domains. To achieve true interoperability, these disparate structures must be aligned through ontology mappings—sets of rules that define relationships between concepts across different schemas.
Manual mapping is prohibitively expensive for large-scale applications, leading to the rise of automated "matchers." While these tools are necessary, they frequently produce outputs characterized by uncertainty and logical inconsistency. This paper proposes a formal logic-based framework designed to merge these "messy" mappings, providing a mathematically rigorous method for handling confidence levels and resolving contradictions.
The Limitations of Automated Mapping
Automated matching systems, such as Falcon, HMatch, and COMA++, typically suffer from three primary deficiencies that hinder their utility in formal reasoning systems:
- Uncertainty: Matchers generate confidence scores rather than absolute certainties, yet standard web ontology languages like OWL are not designed to process probabilistic data.
- Inconsistency: Different tools often suggest mappings that logically contradict one another or violate the internal constraints of the underlying ontologies.
- Oversimplification: Most tools focus on simple 1-to-1 equivalences, failing to capture the complex, conditional relationships required for sophisticated data integration.
A Hybrid Logical Approach
To address these challenges, the proposed framework merges two pillars of computational logic: Description Logics (DL), which provide the formal basis for web ontologies, and Logic Programming (LP), which excels at representing complex rules and nonmonotonic reasoning.
The architecture relies on several key technical components:
- Tightly Integrated DL Programs: Unlike previous "loose" integrations that kept rules and ontologies in separate silos, this framework allows ontology concepts and roles to appear directly within the heads and bodies of logic rules. This integration significantly increases the expressivity of the mappings.
- Independent Choice Logic (ICL): By incorporating Poole’s ICL, the framework can represent a "choice space" consisting of various possible worlds, each assigned a specific probability. This allows the system to treat matcher confidence scores as probabilities of correctness.
- Answer Set Semantics: The system utilizes the FLP-reduct to manage "nonmonotonic negation" (the ability to assume something is false if it cannot be proven true) and disjunctions (representing "either/or" scenarios).
Probabilistic Integration and Conflict Resolution
One of the framework's core strengths is its ability to interpret a matcher’s confidence score ($n$) as the probability that a specific mapping is valid. By assuming that different matchers operate independently, the framework can apply standard probability theory to combine their results. For example, if two independent tools identify the same mapping, the system mathematically increases the overall confidence in that relationship.
Furthermore, the framework utilizes disjunction to handle logical conflicts. If one matcher categorizes a publication as a "Long Paper" while another labels it a "Short Paper," the framework can represent this as a disjunction ($Long \lor Short \leftarrow Source$). This prevents the entire reasoning system from collapsing due to a logical contradiction, instead maintaining a consistent probabilistic model where both possibilities are weighed.
Evaluation and Tractability
The authors validated the framework using real-world data from the Ontology Alignment Evaluation Initiative (OAEI) 2006, specifically the Benchmark and Conference datasets. These datasets include OWL ontologies for scientific publications and conference management.
The evaluation demonstrated that the framework could successfully merge conflicting rules from systems like HMatch and Falcon. In one instance, the framework calculated a combined mapping probability of 0.9892 by integrating results from two independent matchers.
Crucially, the paper addresses the issue of computational complexity. The authors provide a proof of tractability, showing that if the underlying ontologies are restricted to DL-Lite (a streamlined version of description logic), reasoning within this complex probabilistic framework remains polynomial relative to the data size. This finding suggests that the approach is viable for practical applications.
Conclusion
This research provides a robust theoretical foundation for representing and merging uncertain ontology mappings. By extending both standard logic programming and description logics, the authors have created a decidable framework capable of resolving the inherent noise of automated matching. While future work is required to bridge the gap between this theoretical model and a full-scale implementation—and to address scalability for the most expressive versions of OWL—this framework represents a significant step toward more resilient and intelligent data integration on the Semantic Web.
Summary updated