A Contextual Clustering Approach for Theory Manipulation of RKF Knowledge Bases

   Capabilities Being Demonstrated:
  • Organization of large ontologies
  • Detection of infelicitous usage patterns
  • Detection of componentizable regions
  • Template formation for high-level knowledge entry

DARPA's Rapid Knowledge Formation project's central objective is to enable distributed teams of subject mater experts (SMEs) to enter and modify knowledge directly and easily, without need for specialized training in knowledge representation acquisition, or manipulation. Multi-ViewPoint-Clustering Analysis (MVP-CA) technology is being used currently to bridge the gap between the formal logic of machines and the "not-so-formal" logic of humans by providing a support infrastructure for the comprehension, manipulation and utilization of the knowledge in the knowledge base.

We are accomplishing this objective by clustering facts, ontologies and implication rules in existing RKF knowledge bases. Clustering axioms provides a means for exposing the various knowledge base concepts in the context of their actual usage as opposed to exposing them in the context of their declaration. Besides providing a comprehension aid base for the various knowledge bases, clustering also helps identify the missing or misconceived knowledge pieces so that necessary on-time "knowledge repair" can be performed, to render the system more robust and reliable. In addition it allows the knowledge to be filtered, manipulated, summarized, and structured for efficient usage in the long-term.

In the RKF project, we are focusing on clustering very large knowledge bases, such as, Cycorp's Cyc, which are too large and complex to be understood through manual inspection alone. Pragati's MVP-CA tool is helping the Cyc team in RKF by primarily extracting the various subdomains in Cyc knowledge bases, then showing the interrelationships of concepts as they exist in the ontology and the axioms, discovering infelicitious usage patterns in the knowledge base, and aiding the higher level knowledge entry in the system through formation of templates and identification of reusable components. A similar approach is being utilized to cluster KM axioms being formulated through a component-based approach by the SRI team in the RKF project.

A parallel effort is also underway to use clustering in the evaluation of these multiply-authored knowledge bases. The intent is to expose knowledge base quality isues such as:
  • multiple versions of similar concepts across the knowledge bases,
  • insufficient/absence of utilization of existing concepts
  • insufficient faceting in knowledge representation
  • overspecialization of terms in the system
  • which precludes reuse of concepts, and exposes compositionality and maintainability issues.
Click here to access password protected Pragati's RKF presentations.

« Back | Top
2007 Pragati Synergetic Research, Inc. Research Focus | Products | Projects | Publications | Market Sectors | About Us