Knowledge Engineering using The Weak-Transitive-Closure algorithm -- A case study of The University of Oslo's IT system

Abstract

The complexity of human-computer systems nowadays requires the aid of manageable and simplified, machine-readable representation of those systems. In addition, the need to accommodate appropriate and mandatory changes to legacy systems is an inherent challenge in system administration tasks. Thus, a state-of-the-art knowledge management system must be capable of balancing both these needs.

A reasoning method that works via a simplified calculus of facts and rules was suggested by professors Alva L. Couch of Tufts University and Mark Burgess of Oslo University College. The name `weak-transitive-closure algorithm' or the WTC algorithm for short, is adopted here for this method owing to the key technique applied in the system. This method visualizes complex systems as directed graphs, and applies graph theory techniques for inference of causal dependence between components of the system.

This thesis is an investigative study of the application of the WTC algorithm, using University of Oslo's IT system as the application domain. A topic map representation already exists for this system. The application of the WTC algorithm requires a thorough study of the problem domain, so as to construct a knowledge base and a set of rules as the embodiment of the abstractions, modeling and representation of the system. These sets of facts and rules are then input to a prototype engine, which uses them for supplying answers to queries about the system. In contrast, the topic map representation of the system is as undirected graph, comprised of discrete components called topics connected by the edges known as associations. Evaluation of the two representations at various levels is done thoroughly, since both enforce some constraints on how to model and represent the system. In the process, the new opportunities of the weak-transitive-closure algorithm in supplementing and/or replacing the topic map representation are investigated.

We demonstrate that the WTC algorithm has the advantage of discovering connections with specific properties, by generating the paths automatically, which is more optimized for troubleshooting. In addition, the WTC algorithm's presentation is more suitable for learning about legacy systems.