Tuesday, 5/22/2007
4:00 PM - 4:30 PM
Level: Technical - Intermediate

The Semantic Web is based on the model of “distributed syndication” of knowledge. This necessitates usage of good tools for knowledge engineers to “comparison shop” for relevant concepts on the web so that interoperability across disparate knowledge sources can be enabled efficiently. Unfortunately, no good tools for ontology mapping exist to date even though a lot of research has been done on mapping algorithms. This is because the real-world of semantic web ontologies is often messy and the cognitive aspects of deploying mapping algorithms have been largely underestimated and mostly ignored. Some visualization tools are able to expose ontological terms and their declared relationships, but when it comes to providing mapping support for comparing multiple ontologies, they mostly fail in flagging the related regions across various sources. A cognitive support infrastructure to aid reliable and efficient mapping is very much needed. The Expozé tool suite utilizes a clustering-based approach to expose various types of inter-concept relationships from different knowledge sources on the web. Clustering can statically partition the ontologies/knowledge bases thus reducing information overload for knowledge engineers. It significantly lowers the knowledge comprehension barrier by exposing contextually related conceptual islands in the various knowledge sources. In addition, clustering reveals intuitive information about these knowledge bases and ontologies that formal analysis cannot easily achieve. In particular, through Expozé, we can expose similarities and/or differences at the modeling as well as at the terminological level. In addition to exposing the “traditional” relationships such as equivalence and subsumption, etc. our heuristics-based algorithms, also exposes other types of subtle relationships such as inverse, cliché’s templates and others. Our approach provides a unique way for users to access concepts in existing web content and gain insight into their ontological utility. The clustering tool is domain and representation independent. In our talk we will be presenting an overview of our technology together with some real use cases from knowledge bases on the web that highlight the subtleties in the exposed results.

Mala Mehrotra is the founder and President of Pragati Synergetic Research Inc., a leading technology-provider for analysis of information systems, such as ontologies, rule-bases and data base schemas. She has spearheaded several projects since its founding in 1993 including SBIR awards and other contracts from agencies like from US Air Force, Navy, Intelligence Community, NASA, DARPA, DOT and NSF. She was part of DARPA's Rapid Knowledge Formation program and is currently participating in W3C's RIF working group. She has published several papers in the areas of knowledge capture, reuse, mapping and quality assurance.