面向代码坏味检测的阈值动态优化方法

发布时间：2018-01-30 05:44

本文关键词： 软件重构代码坏味反馈控制坏味检测　出处：《北京理工大学》2016年硕士论文　论文类型：学位论文

【摘要】：软件重构是通过调整软件内部结构来提高软件质量的一种技术。为了判定哪里需要重构,人们提出了代码坏味的概念用于表示设计不合理的代码。但代码坏味的手工检测比较困难也需要消耗大量宝贵的人力资源,因此人们提出了各种各样的代码坏味检测方法与检测工具。考虑到软件重构的主观性,这些算法通常会给程序员预留一些阈值进行个性化的定制。一方面是因为代码坏味检测通常是很主观的并且与应用程序相关。比如,长方法的检测,代码坏味检测算法需要知道多长才是真正意义上的长。另一方面,程序员有自己不同的工作进度安排和软件质量需求,他们需要根据自己的实际情况来个性化的定制相关阈值。然而,大部分的程序员并不了解阈值设定和性能表现之间精确的数量关系,因此要他们手动地优化这些阈值非常困难。为此,本文提出了一种面向代码坏味检测的阈值动态优化方法。该方法根据程序员的反馈自动优化坏味检测中的阈值设置。具体优化过程包括如下几个步骤。首先,程序员根据自己的工作进度安排和代码质量需求设置一个目标查准率。然后,根据程序员的反馈,阈值优化算法在保证查准率不小于目标查准率的情况下,自动的搜索最优的阈值设定使得查全率最大化。在5个不同类型的开源项目上对4种常见的代码坏味进行了实验验证,结果表明,该方法比现有方法更为有效,实际查准率与目标查准率之间的平均差距降低了80%;并且初始阈值的选择对方法几乎没有影响。因此该方法是十分有效的。
[Abstract]:Software refactoring is a technique to improve software quality by adjusting the internal structure of software. People put forward the concept of bad taste of code to represent the code with unreasonable design, but the manual detection of bad taste of code is more difficult and needs to consume a lot of valuable human resources. Therefore, people put forward a variety of code bad taste detection methods and detection tools, considering the subjectivity of software refactoring. These algorithms usually allow programmers some threshold for personalized customization, on the one hand, because code bad taste detection is usually subjective and relevant to the application, for example, long method detection. Code bad taste detection algorithms need to know how long is really long. On the other hand, programmers have their own different schedule and software quality requirements. They need to customize the relevant thresholds according to their actual situation. However, most programmers do not understand the precise quantitative relationship between threshold setting and performance. So it's very difficult to ask them to manually optimize these thresholds. In this paper, a threshold dynamic optimization method for code bad taste detection is proposed. The method automatically optimizes threshold setting in bad taste detection according to programmer's feedback. The specific optimization process includes the following steps. The programmer sets a target precision rate according to his work schedule and code quality requirements. Then, according to the feedback of the programmer, the threshold optimization algorithm ensures that the precision rate is not less than the target precision rate. Automatic search optimal threshold sets maximize recall. Four common code bad smells are tested on five different open source projects. The results show that the proposed method is more effective than the existing methods. The average gap between the actual precision rate and the target precision rate is reduced by 80%; And the choice of initial threshold has little effect on the method, so this method is very effective.
【学位授予单位】：北京理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP311.1

【相似文献】