Identifying the severity of technical debt issues based on semantic and structural information

Technical debt (TD) refers to the phenomenon that developers choose a compromise solution from a short-term benefit perspective during design or architecture selection. TD-related issues, such as code smells, may have a critical impact on important non-functional requirements. Different severity levels of TD issues require different measures to be taken by developers in the future. Existing studies mainly focus on detecting TD in software projects through source code or comments, but usually ignore the severity degree of TD issues. As a matter of fact, it is very important to identify the severity of TD issues and clarify which TD should be prioritized. In this paper, we propose an approach that combines the semantic and structural information of the code snippets to identify their severity at method level. In the approach, we first transform each method affected by TD issues into an abstract syntax tree (AST) and use the paths in the AST to represent its semantic information. Then, we extract different code metrics to measure the size, coupling, and complexity of methods affected by TD issues to represent their structural information. Finally, we build a stacking ensemble model to identify the severity of TD issues by using Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) for the base classifiers and Support Vector Machine (SVM) for the meta-classifier. The evaluation results on the real dataset show that our approach achieves 65.77% in terms of precision, 68.18% in terms of recall, and 65.84% in terms of F1-score on average. In addition, the experimental results also demonstrate that the strategy of combining the semantic and structural information of code snippets is effective in improving the effectiveness of our approach.