基于关键类判定的代码提交理解辅助方法

发布时间：2018-07-04 17:16

本文选题：代码修改 + 代码修改理解　；参考：《软件学报》2017年06期

【摘要】：软件代码提交是最重要的软件版本演化数据之一,被广泛应用于软件审查和软件理解中.对于程序员,提交的理解难度随着受影响的类数量、修改的代码量的增加而增加.通过对大量数据的分析发现:识别出提交中核心的修改类(关键类)以及为了完成这个核心修改所进行的依赖性改动的类(非关键类),能够辅助代码提交的理解.受机器学习技术在分类领域有效性的启发,提出一种基于机器学习的关键类识别方法,将判定提交中的关键类建模为二分类问题(即关键和非关键类),从软件演化过程中产生的海量提交数据中抽取可判别性特征来度量类的关键性.在多个数据集上的实验结果表明:该方法判定关键类的综合准确率达到了87%;相比于开发人员直接理解提交,使用关键类信息提示来辅助理解提交,能够显著提高开发人员的效率和正确率.
[Abstract]:Software code submission is one of the most important software version evolution data, which is widely used in software review and software understanding. For programmers, the difficulty of understanding submission increases with the number of affected classes and the amount of code modified. Through the analysis of a large number of data, it is found that the core modification class (key class) in the submission and the dependent change class (non-critical class) in order to complete the core modification can assist in the understanding of code submission. Inspired by the effectiveness of machine learning technology in classification field, a key class recognition method based on machine learning is proposed. The key class in decision submission is modeled as a two-class problem (i.e. key and non-critical classes), and the discriminability features are extracted from the massive submitted data generated in the process of software evolution to measure the criticality of the class. The experimental results on several data sets show that the synthetic accuracy of this method is 87%, compared with the developer's direct understanding of submission, the key class information hint is used to assist in understanding the submission. It can significantly improve the developer's efficiency and accuracy.
【作者单位】：中山大学数据科学与计算机学院;国家数字家庭工程技术研究中心;中山大学先进技术研究院;北京大学信息科学技术学院软件研究所;高可信软件技术教育部重点实验室(北京大学);
【基金】：NSFC-广东联合基金(U1201252) 国家重点研发计划(2016YFB1000101) 国家自然科学基金(61672545,61672045) 广东科技计划(2015B040403005)~~
【分类号】：TP181;TP311.5

【相似文献】