Dsn2Code: An automated approach for similarity-based Software Architecture selection for Code reuse

Code reuse makes a significant impact on the development process by reducing the development time and increasing the quality of the product. Choosing a relevant source code in design phase is challenging task using an architecture. In this study, we propose a framework for suggesting relevant software design from a local repository based on a target software architecture. Our hypothesis in this study is that relevant architectures hold relevant source codes. The main goal of the framework is to increase the effectiveness of software development by empowering programmers to find and reuse existing designs and their source code which are appropriate for their projects. The framework analyzes and suggests designs that closely match the target architecture using a combination of structural similarity, and semantic analysis approaches. The framework is based on the information extracted from the software design. The extracted information can be real dictionary words or composed words or abbreviated terms. Another challenging task was finding real dictionary words from composed and abbreviated terms. We ranked the architectures by finding the real dictionary words from the terms and assigning scores using spaCy pre-trained model based on how well they matched the information in the current software design. The software design and its source code will be suggested based on the result of the pre-trained model spaCy. Evaluation of our developed framework resulted in approximately 85% accuracy in identifier splitting and 75% accuracy in finding real dictionary words from abbreviated terms by applying edit distance Jaccard coefficient. Overall performance of the proposed model was approximately 85%.