A Graph-Based Approach to Model Management

A graph-based framework for model management system design is proposed in this paper. The framework applies graph theory to the development of a knowledge-based model management system, which has the capability of integrating existing models in the model base to support ad hoc decision making. In other words, models in the model base are not only stand-alone models but also building blocks for creating integrated models. This guarantees effective utilization of developed models and promises future development of an automated modeling system. In the framework, nodes and edges are used to represent sets of data attributes and sets of functions for converting a set of data from one format to another respectively. A basic model is defined as a combination of two nodes, one input node and one output nod2, and an edge connecting the two nodes. A model graph, which is composed of basic models, is a graph representing all possible alternatives for producing the requested information. Each path in a model graph is a model for producing the information. If the path includes more than one basic model, it represents an integrated model. Based on the graphical representation. an inference mechanism for model integration and strategies for model selection are presented. INTRODUCTION This article discusses the application of graph theory to model management system design. A graph-based framework and mechanisms for inGraphs are powerful tools for constructing tegrating models are developed. Because models models and solving problems having to do with are knowledge intensive and composed of many discrete arrangement of objects. Because of functions for converting data, the development their elegance and simplicity, graphs have been of model management systems has traditionally applied in many areas, including economics, been considered as a difficult research problem. operations research, physics, cybernetics, comSuccessful implementation of the framework puter science, and other areas in engineering has not only significantly contributed to our (Busacker and Saaty, 1965). Based on simple knowledge of model management but also idea of nodes interconnected by edges, graph promised more effective use of computerized theory combines these basic ingredients into a models. rich diversity of forms with flexible properties and has contributed to the analysis of a wide variety of systems, such as network flow probThe remainder of this paper is organized as follems, markov chains, and PERT and related lows: first, background of model management techniques in operations research. systems will be briefly described. Then, the