基于RNN进行面向主题的特征定位方法

TOPIC ORIENTED FEATURE LOCALIZATION METHOD BASED ON RNN

软件特征定位是软件演化活动得以顺利展开的前提条件。当前特征定位研究的性能仍有待于进一步提高。为了获得较好的性能,在文件夹粒度上获取主题知识,将系统中同一个文件夹下的所有类(class)划分为同一个主题知识类,提出利用深度学习算法——循环神经网络RNN(Recurrent Neural Networks)进行面向主题的特征定位。同时,在该方法的基础上提出了一种改进的模型。为了使实验结果更具现实意义,与基线方法和其他一些方法相比,将实验数据从10组提高到531组和将检索率从15%缩小到10%,即使在这种情况下,所获得的实验结果,无论是从正面与基线方法相比还是从侧面与目前的一些特征定位方法相比,该方法都获得了不错的性能。

Software feature localization is a prerequisite for the smooth development of software evolution. The performance of the current feature location study still needs to be further improved. In order to obtain better performance, get the subject knowledge in the folder granularity was gotten. All the classes under a folder in the system were divided into the same subject knowledge class, This paper proposed a topic-oriented feature locating using depth learning algorithm-Recurrent Neural Networks（ RNN） . At the same time, an improved model was proposed based on this method. In order to make the experimental results more realistic, compared with the baseline method and other methods, this article will test data from 10 to 531 group and the retrieval rate from 15% to 10% . The experimental results show that this method has better performance than either the baseline method or the feature orientation method.