A Rule-based Framework for Enhancing Architectural Decision Guidance

Abstract

Architectural decision-making is a non-trivial process for architects in software development projects. In many cases, such a process starts by identifying architectural issues that an architect should make decisions about. In the second step, the architects explore available alternatives to solve the architectural issues. In the final step, the architects choose one of the candidate alternatives for each issue, based on the decision drivers. Notable progress has been made to assist practitioners in choosing one alternative among the possible alternatives. Several methods and tools have been developed for documenting the rationale and the outcome of the decision-making process. There is, nevertheless, little research focusing on identifying architectural issues that are eligible for a particular project. In the absence of systematic methods of identifying architectural issues, practitioners mainly start their architectural decision-making process based on intuition and prior experience, which may be insufficient due to cognitive biases. We have investigated the industrial context to understand the attitudes and challenges of large-scale enterprises in making and reusing architectural decisions. Then, we have reviewed the literature to identify the gap in developing architectural knowledge about the past into architectural decision guidance for the future. Afterwards, we have tackled the problem of enhancing architectural guidance by developing a framework called Semi Automated Design Guidance Enhancer (SADGE). SADGE extracts architectural issues from project documents and domain literature by applying natural language processing (NLP). This encourages practitioners to identify more architectural issues in the early phases of their projects, making them more prepared for the later phases, when changing and/or refactoring the architecture is more costly. Finally, we have evaluated the framework by conducting a case study on project documents and running experiments with IT students and expert IT architects. The results of the evaluation show that SADGE extracts architectural issues with a significant recall while reducing the manual knowledge processing effort notably. The evaluation also reveals that the experts believe that the framework can be very helpful for them to either reduce the amount of text to read, or to identify hot spots in their documents that need extra attention. The main contributions of this thesis are: C1 An overview of the state-of-the-art and state-of-the-practice in making and reusing architectural decisions. C2 A rule-based framework for developing architectural knowledge in project documents and domain literature into architectural decision guidance. C3 Results of empirical evaluation of developing architectural decision guidance by employing a rule-based framework.