Knowledge representation and information extraction for analysing architectural patterns

Abstract

Today, many software architecture design methods consider the use of architectural patterns as a fundamental design concept. When making an effective pattern selection, software architects must consider, among other aspects, its impact on promoting or inhibiting quality attributes. However, for inexperienced architects, this task often requires significant time and effort. Some reasons of the former include: the number of existing patterns, the emergence of new patterns, the heterogeneity in the natural language descriptions used to define them and the lack of tools for automatic pattern analysis. In this paper we describe an approach, based on knowledge representation and information extraction, for analysing architectural pattern descriptions with respect to specific quality attributes. The approach is automated by computable model that works as a prototype tool. We focus on the performance quality attribute and, by performing experiments on a corpus of patterns with forty-five architects of varying levels of experience, demonstrate that the proposed approach increases recall and reduces analysis time compared to manual analysis.