BUILDING AN EXPERT SYSTEM FROM TEXT - A CASE-STUDY

A major bottleneck in the construction of expert systems has traditionally been the solicitation and formalization of expertise from the 'human expert'. This study introduces a model for building an expert system which relies on 'text' as the source of knowledge. The model is introduced via a case study. The case study is the building of a nurse expert system designed to replicate the medical diagnostic activities of professional nurses. The source of expertise was a nurse diagnoses text. The approach of the model is to first identify two categories of objects (e.g. symptoms and diagnoses) from the text. Second, through a process referred to as cover and differentiation - proposed by Eshelman and McDermott, but not previously applied to text-based knowledge - eliminate unnecessary objects. Finally, through a process referred to as enumeration, the model will allow for the dynamic reconstruction of decision trees. This property of dynamic reconstruction is essential to the fine tuning of the text-based expert system. Fine tuning of an expert system is usually accomplished through a series of 'what if' exercises between the knowledge engineer and the human expert. The purpose of enumeration is not to replicate the 'what if' analysis between the engineer and the expert, but rather to replicate the outcomes of such an exchange.