TRELLIS:

Capturing and Exploiting Semantic Relationships

for Information and Knowledge Management

Yolanda Gil (PI), Jim Blythe, Jihie Kim, and Larry Kite

USC Information Sciences Institute

In a world of overwhelming on-line information access and global communications, more and more people are asked to provide faster and more accurate answers based on up-to-date knowledge that is increasingly more disseminated in vast amounts of information sources. The problem is especially acute in the world of intelligence analysis, the proposed application area for this work.

There are two key research objectives in this work. First, we propose to develop techniques to capture and exploit semantic interrelationships among information items. Our approach will be to use a semantic markup vocabulary that will enable users to specify semantic annotations not only the information items themselves but also the links that users make to relate individual items. We will provide a core vocabulary that will contain both general and domain-specific terms, and that will be extensible by users. We will develop tools that will analyze and exploit these annotations to support users in further analysis, sharing, and integration.

A second objective of this research is to support users in creating new knowledge fragments from raw information sources and from other fragments. The key to our approach is to use the semantic annotations to capture the derivation and rationale of their answers to stated questions as they progressively generate new knowledge based on their expertise and viewpoint. Capturing this information results in significant added value to the original raw information sources. We will support users to highlight key salient information from large reports and documents, to add new intermediate knowledge fragments based on their analysis and integration of existing information, and to finally put together these fragmented pieces into an overall picture of the situation.

This work is motivated by our previous research on knowledge acquisition within the EXPECT project. In order for non-programmers to add knowledge into a system, they need to be guided step by step through the modelling and knowledge representation decisions that knowledge engineers normally do. Users need to be guided through several steps: 1) collecting relevant documents and other information and data sources, 2) analyzing, grouping, and indexing related information, 3) relating the information into structured and consistent form, and 4) formalizing the knowledge into a logic formalism that supports automated inference and reasoning. In our approach, a user could use semantic annotations to specify how each piece of knowledge comes about as they follow each of these steps. The results of each step would remain part of the knowledge base, so the rationale for each piece of knowledge in the system is captured. This approach would be very useful to maintain the knowledge base and to integrate it with other reasoning modules, since the source and rationale for each piece of knowledge will be available in the knowledge base itself instead of disappearing with the knowledge engineers that created it.

The ultimate scientific goal of the project is to contribute to the vision of a Sematic Web that has been put forward by the World Wide Web consortium. To this end, we will collaborate with the DARPA Agent Markup Language (DAML) program and the DAML extended community.

Our approach has significant implications and benefits for the management of knowledge assets that many companies and government institutions are beginning to practice. Making documents available on-line and providing indexing and keyword search are a good first step, but our approach would support more ambitious information processing and management than ever before. By providing increasingly more structure to on-line information, as well as means to customize the way the information is organized, our approach would enable the development of intelligent information management systems that can process and retrieve information in ways specified by the end users themselves. This would result in a new generation of knowledge management, sharing, and dissemination systems.

This work is funded by the Air Force Office of Scientific Research (AFOSR) through funding from the National Reconnaissance Office (NRO).

Relevant links:

Back to the EXPECT group pages.

This page is under construction

Last updated: July 2000