Using domain knowledge to select solutions in abductive diagnosis

This paper presents a novel extension to abductive reasoning in causal nets, namely the use of domain knowledge to select among alternative diagnoses. We describe how preferences among multiple causes of a given state can be expressed in terms of causal nets, and how these preferences can be used to select among alternative diagnoses. We investigate this new extension by proving a number of properties, and show how our preference scheme interacts with conventional ways of choosing among competing diagnoses. Our extension increases the expressive power of causal nets, enjoys a number of desirable properties, and compares favourably with existing proposals for expressing preferential knowledge in causal nets.