Agent-based ontology mapping and integration towards interoperability

Li Li, Yun YangExpert Syst. J. Knowl. Eng 2008 · DOI

TL;DR

This paper introduces a novel agent-based framework for managing ontologies in dynamic, distributed environments. It addresses the critical challenge of achieving interoperability between diverse ontologies by proposing flexible mechanisms for ontology mapping and integration, demonstrated through a practical beer ontology example.

What Problem Does This Paper Address?

In today's interconnected world, data is increasingly distributed across various sources, each potentially using its own ontology to define concepts and relationships. This heterogeneity poses a significant challenge for data integration and system interoperability. Manually mapping and integrating these ontologies is a time-consuming, expensive, and often infeasible task, especially in dynamic environments where ontologies evolve. Existing approaches often lack the flexibility and extensibility needed to handle these complexities. This paper tackles the problem of achieving semantic interoperability by developing an agent-based system that can dynamically map and integrate ontologies.

Methods

The core of the paper is a novel agent-based framework designed for ontology management. This framework utilizes a multi-agent system (MAS) where specialized agents collaborate to achieve ontology mapping and integration. Key agents include:

  • Ontology Agents (OAs): Act on behalf of specific ontologies, providing access to their structure and content.
  • Mapping Agents (MAs): Responsible for identifying semantic relationships (equivalence and inclusion) between concepts in different ontologies. They employ a set of predefined rules based on labels, synonyms, and attributes.
  • Integration Agents (InAs): Combine the results of ontology mapping to create a unified ontology. They count concept occurrences, filter unexpected concepts, and ensure consistency.
  • Functionary Agents (FAs): A group of agents providing specific functionalities like thesaurus similarity, consistency checking, and querying.
  • User Agent (UA), Interface Agent (IA): Handle user interaction and communication brokering.

The framework emphasizes flexibility and extensibility, allowing agents to be added or removed dynamically. The mapping process involves comparing concepts based on syntactic similarity and predefined rules, while integration focuses on combining these mappings to form a coherent, new ontology.

Datasets

The paper uses four different beer ontologies for demonstration and evaluation. These ontologies are derived from various sources:

  • A DAML ontology library.
  • WordNet.
  • A Belgian beer website.
  • An Australian beer types ontology based on web information.

These datasets were chosen to represent diverse perspectives on the "beer" domain and to illustrate the challenges of integrating ontologies with different structures and conceptualizations.

Key Insights and Ideas

  • Agent-based approach is crucial for dynamic environments: The paper strongly advocates for an agent-based approach due to its inherent flexibility, interactivity, and ability to handle dynamic changes in ontologies and environments, which are critical for real-world applications.
  • Separation of concerns through specialized agents: The framework's success hinges on the clear division of labor among specialized agents (e.g., OAs for ontology access, MAs for mapping, InAs for integration), leading to a modular and extensible system.
  • Mapping equivalence is robust, inclusion is nuanced: The evaluation shows that semantic equivalence mapping is highly accurate, likely due to the effectiveness of syntactic similarity measures. However, mapping inclusion is more challenging and sensitive to the conceptualization differences between ontologies.
  • Consistency checking is vital for integration: The integration process relies heavily on consistency checking to refine the results of mapping, particularly for inclusive relations, ensuring the final integrated ontology is coherent.

Key Results

The paper demonstrates the effectiveness of its agent-based framework through a prototype applied to beer ontologies.

  • Ontology Mapping: The evaluation shows that the MAS mapping mechanisms achieve perfect accuracy for equivalent relations, matching human user expectations. For inclusive relations, the results are good but show variations depending on the specific ontologies being compared, highlighting the impact of differing conceptualizations.
  • Ontology Integration: The integration process, aided by consistency checking, successfully produces a new ontology. The results indicate that after mapping and integration, the ontology is accurate for both equivalent and inclusive relations. The prototype successfully demonstrates the creation of an integrated ontology in RDF(s) format.

Limitations and Future Work

The authors acknowledge several areas for future work:

  • Extending agent capabilities: They plan to create more functionary agents, such as planning agents, to enhance the system's capabilities.
  • Advanced semantic matching: Investigating ontology mapping that considers axioms and treats ontologies as algebraic abstractions is a key future direction. This aims to handle more complex relations beyond simple taxonomic ones.
  • Process algebra for agent interactions: The authors intend to use process algebra to support agent interactions at a higher level of abstraction, aiming for more flexible and abstract agent communication.
  • Handling polysemous words: While not explicitly a limitation discussed, the assumption of no polysemous words in ontologies suggests that handling ambiguity in concept meaning could be a future enhancement.

Summary updated

BibTeX
@article{li2008agent,
  author = {Li Li and Yun Yang},
  title = {{Agent-based ontology mapping and integration towards interoperability}},
  journal = {Expert Syst. J. Knowl. Eng.},
  year = {2008},
  doi = {10.1111/j.1468-0394.2008.00460.x},
}
Figure