基于机器学习的源程序自动标注方法  

AUTOMATIC ANNOTATIONS METHOD OF SOURCE PROGRAM BASED ON MACHINE LEARNING

作　　者：王瑞 吴军华 Wang Rui;Wu Junhua(College of Computer Science and Technology,Nanjing Tech University,Nanjing 211816,Jiangsu,China)

机构地区：[1]南京工业大学计算机科学与技术学院,江苏南京211816

出　　处：《计算机应用与软件》2021年第9期16-20,104,共6页Computer Applications and Software

摘　　要：程序注释是软件开发过程中不可缺少的工作。大量的Java程序需要准确的注释来提高程序的可维护性。对自动标注方法进行研究,分析现存方法存在的问题,为了改善注释的效率,提出一种基于机器学习的程序注释自动标注方法用于Java程序的自动标注。方法主要分为两个部分:数据的预处理和机器学习模型。数据预处理采用双编码器对程序进行处理,通过已训练的GRU神经网络模型对未注释的Java程序进行自动标注。实验表明该自动标注方法在准确率、时间等性能方面都有显著的提升,提高了源代码的可读性。Program comments are an indispensable part of the software development process.A large number of Java programs require accurate annotations to improve program maintainability.The automatic labeling method is studied and the problems of existing methods are analyzed.To improve the efficiency of annotation,an automatic annotations method of source program is proposed based on machine learning,which is used for automatic annotation of Java programs.It was mainly divided into two parts,data preprocessing and machine learning model.The data preprocessing mainly used a dual encoder to preprocess the program,and then automatically annotated the uncommented Java program through the trained gated recurrent unit(GRU)neural network model.The experimental results show that the validity and time performance of the presented method are significantly improved,and it also improves the readability of the source code.

关 键 词：程序自动标注 JAVA 机器学习 编码器 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]