Date of Award

1988

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Bush Jones

Abstract

Knowledge integration is an important topic for knowledge engineering. In this dissertation, we explore some aspects of knowledge integration, namely, accumulation of scientific knowledge and performing analogical reasoning on the acquired knowledge. Knowledge to be integrated is conveyed by paragraph-like pieces, these pieces will be referred to as documents. By incorporating some results from cognitive science, the Deutsch-Kraft model of information retrieval is extended to a model for knowledge engineering, which integrates acquired knowledge and performs intelligent retrieval. The resulting computer model is termed COGMIR, which stands for a COGnitive Model for Intelligent Retrieval. A scheme, named query invoked memory reorganization, is used in COGMIR for knowledge integration. Unlike some other schemes which realize knowledge integration through subjective understanding by representing new knowledge in terms of existing knowledge, the proposed scheme suggests at storage time only recording the possible connection of knowledge acquired from different documents. The actual binding of the knowledge acquired from different documents is deferred to query time, depending on the actual needs of the query. Therefore, although there is only one way to store knowledge, there are potentially numerous ways to utilize the knowledge. From the classical information retrieval viewpoint, we have extended the original model in the following sense, not only each document be represented as a whole, but also the meaning of each document can be represented. In addition, since facts are constructed from the documents, document retrieval and fact retrieval are treated in a unified way. Moreover, when the requested knowledge is not available, query invoked memory reorganization can generate suggestion based on available knowledge through analogical reasoning. This is done by revising the algorithms developed for document retrieval and fact retrieval, and by incorporating Gentner's structure mapping theory. Analogical reasoning is treated as a natural extension of intelligent retrieval, so that two previously separate research areas are thus combined. A case study is provided to demonstrate the fundamental ideas. All the components are implemented as list structures, which bears an interesting similarity to relational data-bases.

Pages

195

Share

COinS