A conceptual model and an implementation of adaptive decision support systems

Abstract

Adaptation of decision support systems (DSS) is an important issue in DSS research. Previous research into adaptation of DSS has been focused on individual elements of traditional three-component architectures. The dissertation proposes and validates an integrated conceptual model of adaptive decision support systems (ADSS) which adapts its functions, stmcture, and interfaces in order to support the changing needs of decision makers. The conceptual model is a generic architecture for the development of domain-specific ADSS.

Following a unified research methodology, the dissertation specifically addresses the following issues: 1. What is the adaptive behavior of an ADSS? 2. What knowledge and capabilities are needed to embody the identified adaptive behaviors? 3. What architecture is required to support these capabilities?

Based on previous research into adaptivity of information systems, five adaptive behaviors of ADSS are identified. Eight different types of knowledge are recognized to support the five adaptive behaviors. The model of reflexive systems and a framework of decision making organization are used as theoretical foundations for stmcturing various components. The ADSS conceptual model consists of three units at two levels: the meta-level and the basic-level. Two units at the basic-level are user interface unit and problem processing unit, which are responsible for communicating with the user and carrying out the task of decision support, respectively. The meta-level unit has an introspection mechanism and a self-knowledge base, which comprise a controlling unit capable of introspecting the system's knowledge and limitations, and determining an appropriate action to adjust the capabilities of basic-level units.

The notion of software agents was employed to develop a prototype system in order to examine the feasibility ofthe conceptual model. Software agents were organized in the form of a federate agent-based architecture. The field of real estate was used as the problem domain in developing the prototype. Three types of decision tasks were implemented in the prototype system. A panel of three experts knowledgeable about information systems evaluated the prototype system against five major features. The evaluation results validate the feasibility and usability ofthe conceptual model of ADSS.