一种无监督的软件复杂度度量与评估模型  

An unsupervised measurement and evaluation model of software complexity

作　　者：柯文俊 王泊涵 杜泽峰 姜利 缪沛恩 Ke Wenjun;Wang Bohan;Du Zefeng;Jiang Li;Miao Peien(Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190;University of Chinese Academy of Sciences,Beijing 100049;Beijing Institute of Computer Technology and Application,Beijing 100089;Institute of System Engineering,National University of Defense Technology,Changsha 410073;Department of Information Engineering,China University of Geosciences,Beijing 100190;Department of Software Engineering,University of Science and Technology of China,Hefei 230051;Department of Information Engineering,Nanchang University,Nanchang 330000)

机构地区：[1]中国科学院计算技术研究所,北京100190 [2]中国科学院大学,北京100049 [3]北京计算机技术及应用研究所,北京100089 [4]国防科技大学信息工程学院,长沙410073 [5]中国地质大学信息工程学院,北京100190 [6]中国科学技术大学软件工程学院,合肥230051 [7]南昌大学信息工程学院,南昌330000

出　　处：《高技术通讯》2020年第4期333-341,共9页Chinese High Technology Letters

基　　金：国防基础科研(JCKY2016204B503);装发预研基金(61400020103,61400020403)资助项目。

摘　　要：软件复杂度度量作为软件工程的重要组成部分,可为软件的控制和降解、软件测试的资源分配和高质量软件的研制提供支撑。计算机控制软件往往规模复杂,开发、测试和维护难度大,其复杂度的准确度量意义重大。然而,现有方法大多依赖少量度量元或者人工设置各度量元的超参数权重,工作量大且准确度难以保证。本文提出了一种无监督的、自适应的软件复杂度度量算法,对度量元数据进行高斯混合模型(GMM)的概率建模和归一化;通过皮尔逊相关系数过滤度量元间的涌现特征,借助拓扑排序和图论思想,构建度量元的自适应线性加权模型,综合评估软件的复杂度。在3种数据集上的实验表明,本文提出的软件复杂度度量模型在定量和定性分析结果上取得了更好的度量和评估效果,可以有效解决软件的复杂度度量问题。As an important part of software engineering,software complexity measurement supports for software control,degradation and resource allocation in the field of software testing and high-quality software development.Computer control software is often complex and difficult to develop,test and maintain.So it is significant to measure its complexity accurately.However,existing methods often use a small number of metric metadata or set the weights of each metric metadata manually,which makes it difficult to guarantee the accuracy and workload.An unsupervised and adaptive software complexity measurement algorithm is proposed.Firstly,an probability modeling and normalization of each metric metadata are carried out with Gaussian mixture model(GMM).Then,with correlation of each two metric metadata filtered with Pearson correlation coefficient,an adaptive linear weighting model of those metric elements is constructed based on topological sorting and graph theory.Finally,experiments on three data sets show that the proposed complexity measurement model achieves better measurement and evaluation results in both quantitative and qualitative analysis,and measures software complexity effectively.

关 键 词：计算机控制软件 高斯混合模型(GMM) 复杂度度量 顶点表示活动(AOV)网络 概率归一化 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]