CoMMA: Transforming Corporate Memory into a Living Multi-Agent Ecosystem

Executive Summary

TL;DR: CoMMA (Corporate Memory Management through Agents) is an ambitious architectural framework that moves away from static "knowledge silos" toward a dynamic, agent-driven ecosystem. By combining several emerging technologies—JADE for agent orchestration and RDF for semantic metadata—it automates the entire lifecycle of corporate knowledge: from insertion and annotation to proactive "push" delivery.

Positioning: This work is a seminal implementation of FIPA-compliant agents applied to Knowledge Management (KM), representing a transition from simple information retrieval to specialized, autonomous sub-societies that handle enterprise ontologies and user modeling.

The "Static Repository" Bottleneck

In the fast-paced ICT sector, knowledge is not just vast; it is volatile. The paper identifies a critical pain point: most "corporate memories" are graveyard-like databases. They suffer from:

• Information Overload: Users cannot find what they need in massive Intranets.
• Reactive Only: Knowledge remains hidden until someone specifically searches for it.
• Rigidity: Updating the underlying data structure often breaks the retrieval interface.

The authors' insight was to treat the corporate memory as a society, where specialized agents take responsibility for specific domains (like users or documents), allowing the system to grow organically without centralized failure.

Methodology: The Four Sub-Societies

The core of CoMMA is its modularity, achieved through the separation of concerns into four distinct groups of agents:

1. Document & Annotation Sub-society: These act as "Archivists." They don't just store files; they manage the XML/RDF metadata that makes documents machine-understandable.
2. Ontology Sub-society: The "Librarians." They manage the Enterprise and User models, ensuring that a query for "Java" understands the context of "Programming Languages."
3. User Sub-society: The "Personal Assistants." They learn user preferences and adapt the UI, handling the "Push" mechanism to notify employees of relevant new information.
4. Interconnection Sub-society: The "Matchmakers." They facilitate discovery between agents, ensuring a User Agent can find the right Archivist Agent.

Figure 1: While specific diagrams represent the organizational context, the architecture relies on the JADE framework to decouple task logic from communication.

Semantic Intelligence: RDF & CORESE

A critical technical pillar of CoMMA is CORESE (Conceptual Resource Search Engine). Unlike keyword search, CORESE works with RDF annotations. When a user adds a document, the system doesn't just index text; it creates a semantic map.

• Inference Power: Because it uses Conceptual Graphs (CG), the system can "infer" relationships not explicitly stated in the query.
• Machine Learning: User agents track interaction patterns to refine "push" notifications, ensuring the right knowledge reaches the right employee at the right time.

Experimental Validation

The system was deployed in real-world environments to support two primary use cases:

• New Employee Integration: Streamlining the "onboarding" process by pushing relevant corporate culture and technical docs.
• Technology Monitoring: Detecting market and technology shifts and disseminating them across the organization.

Key Findings:

• Modularity: Changing the "Archivist" behavior to support a new database type required zero changes to the "User" or "Ontology" agents.
• FIPA Compliance: Using the JADE framework ensured that the system remains interoperable and follows international standards for agent communication.

Critical Insight & Conclusion

CoMMA’s greatest contribution is the proactive agent. By shifting from a "Pull" model (where users must seek) to a "Push/Pull" hybrid (where the system anticipates), it addresses the human factor of knowledge management—that people often "don't know what they don't know."

Limitations: At the time of publication, the manual overhead of RDF annotation remained a hurdle. Modern iterations would likely replace manual XML tagging with LLM-based auto-tagging.

Future Outlook: The CoMMA architecture laid the groundwork for what we now call "Semantic Desktops" and "Enterprise AI." Its sub-society model is a direct precursor to modern Multi-Agent Systems used in autonomous R&D workflows.