HP Labs Technical Reports
Click here for full text:
Uncertainty Modelling in Diagnostic Systems: An Adaptive Solution
Keyword(s): diagnosis; case-based reasoning; artificial intelligence; information retrieval; knowledge management
Abstract: Uncertainty permeates the entire diagnostic process and its management is a fundamental issue in actual diagnostic systems. The type of information we can model about the context in which a problem occurred is crucial. The main components pictured in the definition of a context are observations (facts) but we argue that data on relevance and confidence may add precious information. Focusing on case-based reasoning (CBR) paradigm, we present a model in which relevance and uncertainty become fundamental and dynamic components of both diagnostic knowledge and processes: fuzzy sets are the theoretic base of the model. A conversational CBR shell implementing nearest- neighbour (NN) retrieval mechanisms has been developed in order to test our proposal in terms of case- retrieval precision and we discuss the results obtained in some experiments. The "knowledge level" impact of our proposal is also discussed.
Back to Index