Knowledge engineering with semantic web technologies for decision support systems based on psychological models of expertise

Abstract

Machines that provide decision support have traditionally used either a representation of human expertise or used mathematical algorithms. Each approach has its own limitations. This study helps to combine both types of decision support system for a single system. However, the focus is on how the machines can formalise and manipulate the human representation of expertise rather than on data processing or machine learning algorithms. It will be based on a system that represents human expertise in a psychological format. The particular decision support system for testing the approach is based on a psychological model of classification that is called the Galatean model of classification. The simple classification problems only require one XML structure to represent each class and the objects to be assigned to it. However, when the classification system is implemented as a decision support system within more complex realworld domains, there may be many variations of the class specification for different types of object to be assigned to the class in different circumstances and by different types of user making the classification decision. All these XML structures will be related to each other in formal ways, based on the original class specification, but managing their relationships and evolution becomes very difficult when the specifications for the XML variants are text-based documents. For dealing with these complexities a knowledge representation needs to be in a format that can be easily understood by human users as well as supporting ongoing knowledge engineering, including evolution and consistency of knowledge. The aim is to explore how semantic web technologies can be employed to help the knowledge engineering process for decision support systems based on human expertise, but deployed in complex domains with variable circumstances. The research evaluated OWL as a suitable vehicle for representing psychological expertise. The task was to see how well it can provide a machine formalism for the knowledge without losing its psychological validity or transparency: that is, the ability of end users to understand the knowledge representation intuitively despite its OWL format. The OWL Galatea model is designed in this study to help in automatic knowledge maintenance, reducing the replication of knowledge with variant uncertainties and support in knowledge engineering processes. The OWL-based approaches used in this model also aid in the adaptive knowledge management. An adaptive assessment questionnaire is an example of it, which is dynamically derived using the users age as the seed for creating the alternative questionnaires. The credibility of the OWL Galatea model is tested by applying it on two extremely different assessment domains (i.e. GRiST and ADVANCE). The conclusions are that OWLbased specifications provide the complementary structures for managing complex knowledge based on human expertise without impeding the end users’ understanding of the knowledgebase. The generic classification model is applicable to many domains and the accompanying OWL specification facilitates its implementations.