SpecRefiner: Assistance for stepwise refinement of software development in the upstream process

Abstract

AbstractA system supporting the refining of an analytical model in the upstream processes of software development has been developed. Compared with the absence of such support, human oversights during system analyses are reduced. Although applications and implementation support tools for downstream processes of software development have been vigorously conducted in the past, since there are many stages in studying the applications of support tools for upstream processes, especially structured analyses and the like, expression of concept models has received almost no support for the analytical processes themselves. SpecRefiner has been constructed as a means of supporting stepwise refinement in structured analyses. SpecRefiner is the system which supports a requested analysis process by storing statements of what is to be performed, in specifications written in natural language and the DFDs (data flow diagrams) accompanying them in a case base, and by detecting similar specifications and DFDs in the stepwise refinement process when performing a newly requested analytical task. The method treats the software development process as consisting of three steps, namely, finding, revising, and refining, and supports each step. It finds a past analyzed case, revises this case, and refines it. In addition, it supports asynchronous collaborative design by retaining design rationale or a design history. Fischer [3, 4] and Nakakoji [10] state that the construction of an application for an area or domain is facilitated by a specialist who constructs a domain‐oriented design support environment by using this tool. In the study by Mizoguchi [9], ontology is organized in area‐independent levels, and design using building blocks corresponding to the ontology is made possible by means of interviews between the SE and the end user based on the ontology. In the investigation of Seta and colleagues [13], the user himself describes the task configuration. The proposed scheme lies between the domain‐oriented design environment of Fischer and the domain‐independent ontology approach of Mizoguchi. The proposed scheme can reduce the cognitive overload of the end user or SE in translating the domain‐oriented work and the implementation‐level work by seamlessly connecting the domain‐oriented knowledge and the domain‐independent ontology or symbolic‐level knowledge. As a result, the proposed scheme can use analytical time effectively by reducing human errors during analyses. © 2003 Wiley Periodicals, Inc. Syst Comp Jpn, 34(6): 32–43, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10019