AEROBA : a control blackboard approach for model formulation

Abstract

During a model formulation process, the control is used to dynamically apply various strategies to the appropriate sources of knowledge for the proper parts of problem. The importance of the control aspect has been emphasized in the literature. However, previous studies in automated model formulation use various AI techniques with pre-defined control strategies and can not provide enough support for simulating the expert's modeling process. In order to develop a computerized system based on the expert's model formulation process, a set of verbalization protocols was collected from the experts in managerial domains. The analysis provides two major findings, opportunism and control features. In terms of opportunism, the expert is understood as using various cognitive panels and frequently changing the focus of attention during the formulation process. During this process, the expert also uses various control features for directing the formulation progress. The protocol analysis provides key inputs to the design of the implementation architecture, AEROBA. Due to the opportunism observed from the expert's protocols, the blackboard architecture is selected for the AEROBA design. In this blackboard architecture, an EER model (Extended Entity Relationship model) is used to define the blackboard data structure, a structure transformation approach based on the EER constructs is used to design the domain knowledge, and finally, a control blackboard approach is used to design the control. The AEROBA control simulates the control aspect observed from the protocol analysis. With the incorporation of the control features, AEROBA supports a diverse reasoning process and hence extends the flexibility of control in automated model formulation. The prototype of the AEROBA system is implemented in Common LISP on a SUN 3/160 workstation. The evaluation of the AEROBA system involves two approaches: validation and verification. Validation is used to check the correctness of the formulations outcomes and the formulation process of the system. Verification is used to determine the usefulness, called the formulation effectiveness, of the system. The evaluation indicates that the AEROBA approach is an appropriate way to realize the abstract control aspect in model formulation into a computerized environment.