ACQUIRING APPLICATION-SPECIFIC KNOWLEDGE DURING DESIGN TO SUPPORT SYSTEMS MAINTENANCE

Abstract

Most large systems development efforts proceed in a top-down fashion where initial specifications and requirements are incorporated into a high-level design, followed by programs based on this design. However, a major part of the software life-cycle effort is devoted to maintenance. While several existing methodologies aid in the initial phases of requirements and specification, they have proven to be of little value for maintenance. Changes in user requirements are often translated directly to the level of code, divorcing it from the high level design it was based on. After a few such changes, the programs may not correspond to any formal high-level design, making subsequent maintenance difficult. We argue that maintenance must be based on the knowledge used in synthesizing the high-level design. This requires a development environment where the knowledge about high-level designs is formally represented, and raises the question about how this knowledge will be acquired by the support environment in the first place. In this paper, we present a model that enables the support environment to acquire design knowledge through "learning by observation" of a designer engaged in specifying a high-level design. The knowledge that the learning system begins with is a generic object for expressing design decisions. Based on the input provided by the designer, and a limited interactive querying process, it constructs and continuously refines a taxonomic classification of application-specific knowledge and rules at an appropriate level of generality that capture the rationale of the design. This knowledge can be used subsequently for maintaining system designs and recognizing design situations similar to the ones it has knowledge about.