Facilitating the use of multi‐attribute utility theory in expert systems:an aid for identifying the right relative importance weights of attributes

Abstract

In general, expert system applications to real cases involve making decisions, i.e. selecting a suitable action among a set of possible alternative actions. A well‐known standard method for modelling decision problems is the so‐called multi‐attribute utility theory (MAUT), a method in which the alternatives are viewed in terms of their attributes. A set of attributes are identified and a specific value and a suitable relative importance weight are assigned to each attribute. However, it is not easy for the expert to quantify the relative importance weight of an attribute: this assignment entails a certain abstraction activity from the expert and, as is well known, experts have difficulty in providing their knowledge in abstract and general terms. In order to overcome this difficulty we propose a method for automatically inferring relative importance weights from a set of specific action sequences. An action sequence is a list in chronological order of the actions executed by the expert when facing specific cases of decision problems. Providing action sequences requires no other effort but remembering specific episodes, and this task is much easier for experts than having to directly provide precise numbers expressing relative importance weights. Moreover in many cases action sequences are already stored in suitable records. Consider, for example, the list of medical tests executed on a given patient, a list included in the patient clinical record stored in the clinical database of a hospital. On the basis of these considerations the proposed method should be useful for designers of expert systems which face problems of choosing the right action among a set of alternative actions.