Аңдатпа
A significant issue in the recent development is the growing concern with the quality of code and low levels of technical debt in the agile software development environments that can lead to an absence of adequate manual reviews of the code. Recent automated review systems primarily rely on the application of static code analysis and they fail to give credit to contextual and semantic information found in commit histories. This paper demonstrates a severe lack in the exploitation of the information stored in natural language and which developers add to their commits to detect the early indications of technical debt trends. To overcome this drawback, a new framework that presents the utilization of Natural Language Processing (NLP) tools to handle commit messages and other metadata to generate an automatic code review and debt analysis is proposed. The methodology is a combination of transformer-based language model, semantic similarity analysis and classification alongside the application of keywords to detect debt-inducing changes. The experimental background is rooted in the examination of publicly available Git repositories comprising annotated examples of technical debt and evaluates performance with the measures of accuracy, recall, and F1-score. The F1-score of the specified method improves the conventional tools of the statical analysis by 18.6% and the precision and the recall values are more than 0.89 and 0.87, respectively. The main component of this work is the availability of environment-sensitive, NLP-based framework, which enhances automated review of the code and the proactive management of technical debt of the agile development cycles.