计算机科学与探索 ›› 2017, Vol. 11 ›› Issue (10): 1599-1608.DOI: 10.3778/j.issn.1673-9418.1609035

• 系统软件与软件工程 • 上一篇    下一篇

利用RNNLM面向主题的特征定位方法

尹春林1,王  炜1,2+,李  彤1,2,何  云1,熊文军1,周小煊1   

  1. 1. 云南大学 软件学院,昆明 650091
    2. 云南省软件工程重点实验室,昆明 650091
  • 出版日期:2017-10-01 发布日期:2017-10-20

Using RNNLM to Conduct Topic Oriented Feature Location Method

YIN Chunlin1, WANG Wei1,2+, LI Tong1,2, HE Yun1, XIONG Wenjun1, ZHOU Xiaoxuan1   

  1. 1. College of Software, Yunnan University, Kunming 650091, China
    2. Key Laboratory for Software Engineering of Yunnan Province, Kunming 650091, China
  • Online:2017-10-01 Published:2017-10-20

摘要: 软件特征定位是软件演化活动顺利展开的保证。基于文本的特征定位方法是目前特征定位研究的一个重要组成部分。当前基于文本的特征定位方法将代码关键词视为独立同分布的个体,忽略了代码间的语境。针对上述问题,基于深度学习语言模型RNNLM(recurrent neural networks language model)提出了一种源代码主题建模方法,并在此基础上实现了特征定位。实验结果表明,与基于LDA(latent Dirichlet allocation)和LSI(latent semantic indexing)的文本特征定位相比较,查准率提高8.61%和2.61%,表明该方法具有较优的查准率。

关键词: 软件特征定位, 软件演化, RNNLM, 主题建模

Abstract: Software feature location is the guarantee of software evolution. Textual based feature location method is an important part of the current research on feature location. The current feature location method based on text regards the code key words as the independent identically distributed individual, and ignores the context of the code. For this question mentioned above, this paper proposes a source code topic modeling method by using deep learning language model RNNLM (recurrent neural networks language model), and realizes localization features on this basis. The experimental results show that the precision rate is improved by 8.61% and 2.61% compared with the text feature localization based on LDA (latent Dirichlet allocation) and LSI (latent semantic indexing), indicating that the method has better precision.

Key words: software feature location, software evolution, RNNLM, topic modeling