An Experience on Reputation Models Interoperability Based on a Functional Ontology

Laurent Vercouter, Sara J. Casare, Jaime Simão Sichman, Anarosa BrandãoIJCAI 2007

Multi-Agent Systems (MAS) often consist of autonomous agents developed by diverse parties using heterogeneous logic. A significant hurdle in these open systems is establishing a common language for "reputation." While reputation is vital for cooperation and trust, different models define it subjectively, contextually, or globally. This paper, originally presented at IJCAI 2007, proposes a framework to bridge these conceptual gaps using a shared functional ontology and structured queries.

The Challenge of Heterogeneous Reputation

In many MAS environments, agents are treated as "black boxes." To interact safely, they must assess the trustworthiness of others. However, a "reputation" value in one system might not translate directly to another. For instance, one model might use a global score, while another relies on subjective direct experience. Historically, platforms like the ART (Agent Reputation and Trust) testbed forced agents to compress complex internal reasoning into a single, lossy numerical scale (0.0 to 1.0). This simplification discards the rich, context-specific data that makes reputation models truly powerful.

A Framework for Interoperability

The authors propose a three-tiered architecture to enable nuanced communication without requiring every agent to understand every other agent's internal logic. The core of this solution is the Mapping Module, which sits between an agent's private reputation model and the external world.

The proposed agent architecture featuring the Reputation Mapping Module as a bridge between internal models and external interactions.
  1. Functional Ontology: Instead of a new model, the framework adopts an existing ontology that identifies universal reputation concepts, such as roles, entities, and types (e.g., direct experience vs. hearsay).
  2. Agent Architecture and Mapping: The Mapping Module acts as a translator, mapping private concepts (like "Image" in a cognitive model) to the shared ontology (e.g., "Primary Reputation").
  3. Structured Communication via nRQL: Rather than exchanging raw numbers, agents use the new Racer Query Language (nRQL) to send structured strings that preserve semantic detail.

Semantic Mapping and nRQL Queries

The power of this approach lies in its ability to handle context. Using nRQL, an agent can query specific attributes. For example, instead of asking for a generic score, an agent can query for the reputation of "John" specifically regarding "Impressionism" paintings. This allows for a "Rosetta Stone" effect: agents only need to map their internal model to the central ontology once to communicate with any other compliant agent.

A conceptual graph illustrating how a query for an individual's skill in a specific context (e.g., John and Impressionism) is structured using the functional ontology.

Implementation and Results

The researchers validated this framework by modifying the ART testbed, a simulation environment where agents act as art critics. By replacing standard numerical messages with nRQL strings, they enabled "finer agent interactions." This integration also improved observability; researchers could now visualize how an agent's perception of another's reputation evolved over time across different categories and eras using new graphical interfaces.

Conclusion and Future Directions

While the framework successfully demonstrates semantic alignment and enhanced expressiveness, it focuses primarily on communication and mapping. Future work is needed to address the "Reasoning Module"—how agents should process this translated information to make decisions. Additionally, while the computational overhead of nRQL is higher than simple floats, the authors argue that the gain in detail and interoperability is a necessary trade-off for complex, open multi-agent systems.

Summary updated

BibTeX
@inproceedings{vercouter2007experience,
  author = {Laurent Vercouter and Sara J. Casare and Jaime Simão Sichman and Anarosa Brandão},
  title = {{An Experience on Reputation Models Interoperability Based on a Functional Ontology}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2007},
}
Figure