Towards Measuring Defect Debt and Developing a Recommender System for Their Prioritization

软件错误是软件维护和开发活动中不可分割的一部分。由于时间有限和预算紧张，软件开发团队无法完全解决问题跟踪系统中存在的所有错误。推迟缺陷修复的短期利益与长期将缺陷保留在系统中的后果之间的权衡被解释为缺陷债务。通常，缺陷债务被定义为在当前版本中发现但没有修复的任何类型的缺陷、失败或错误。在这项提案中，我旨在调查关于缺陷债务的三个研究问题：缺陷债务的本金金额是多少？不良债务的利息金额是多少？在有限的时间内解决错误的最佳顺序是什么，以便将兴趣降至最低？为了回答这些研究问题，我建议将错误分类为容易负债的错误和常规错误。利用规则缺陷来训练基于KNN回归的预测模型，以估计债务本金。我建议使用图论分析来计算利息。最后，建议使用强化学习技术根据错误的债务金额对错误进行优先排序。为了验证我提出的模型的可行性，我将使用从Mozilla Firefox项目和IBM RTC项目收集的错误报告进行实证研究。

Software bugs are an inextricable part of software maintenance and development activities. Due to limited time and tight budget constraints, the software development team is not able to fully resolve all the bugs that exist in the issue tracking system. The trade-off between short-term benefit of postponing the fixing of defect and the consequence of keeping the bug in the system in the long-term is interpreted as a defect debt. Typically, defect debt is defined as any kind of defect, failure or bug that is found but not fixed in the current release. In this proposal, I aim to investigate three research questions regarding the defect debt: What is the principal amount of defect debt? What is the interest amount for defect debt? What is the optimal sequence for resolving the bugs in a limited time in order to minimize the interest? In order to answer these research questions, I propose to categorize the bugs into debt prone bugs and regular bugs. The regular bugs are used to train the prediction model based on KNN-regression for estimating the principal of the debt. I propose to use graph theory analysis in order to calculate the interest. Eventually, reinforcement learning technique is recommended to prioritize the bugs based on their debt amount. In order to validate the feasibility of my proposed model, I will perform an empirical study using the bug reports collected from Mozilla Firefox project and IBM RTC project.