| While information retrieval systems have been relatively successful, their ability to select suitable documents is limited by the keywords found within the documents themselves. Documents contain knowledge that is not adequately described by the imbedded text. In order to exploit this higher level of abstraction of a document, it is necessary to build new information structures that can support Knowledge Acquisition.;The primary emphasis of this thesis was to explore avenues for effective knowledge acquisition from some form of structured repository of information, such as a collection of documents. The model that was chosen for Knowledge Management is based on Conceptual Graph theory, developed by John Sowa. Conceptual Graphs are a form of directed graphs, so efficient graph matching algorithms are required to query the concepts found within the documents. A large graph can be difficult to match, so a decomposition technique, employing two nodes joined by a single arc, or strands, is used to simplify graph encoding and for storing the information in a relational database system.;The Conceptual Graph structure was implemented to support a variety of document repositories and a query development tool was developed to facilitate the conversion of user queries into a format suitable for matching Conceptual Graphs.;It is essential to seek measures or metrics for Knowledge Acquisition in order to compare different techniques or environments. The system was evaluated using extensive performance tests over a wide range of documents, with near linear performance as the number of documents increased. Response time was similar to keyword retrieval times, but with a significant improvement in matching accuracy. Conceptual Graph queries returned an average of 29.4% more documents and eliminated 55.4% of the documents that were not relevant to the query. These encouraging results suggest that imbedding additional knowledge within documents can result in qualitatively better search results with minimal additional overhead to traditional retrieval techniques. |
