基于网络表征学习的软件系统演化分析

Software system evolution analysis based on network representation learning

从包、类和方法3个粒度构建软件元素的网络模型。利用经典的网络表征学习方法Node2vec学习节点特征,并从网络距离、增长特性、更新率、模块度等方面对3个开源软件系统进行演化分析。试验结果表明:3种粒度下的演化特性不尽相同,包粒度下的演化更加稳定且高效;相比先前研究,本研究方法得到的软件演化特性与Lehman定律更契合;当软件系统迭代累计到最大阈值时其体系架构将重新部署,此时软件系统的鲁棒性最差且易产生峭壁。

This research constructed the network model of software elements from package,class,and function granularity.The classical network characterization learning method Node2vec was used to learn node features,and the evolution of three open source software systems was analyzed from the aspects of network distance,growth characteristics,update rate,and modularity.The experimental results showed that:the evolution characteristics of the three particle sizes were different,and the evolution of the package size was more stable and efficient;compared with previous studies,the software evolution characteristics obtained by this method were more consistent with Lehman's law;when the software system iteration accumulated to the maximum threshold,the architecture was redeployed.