代码注释生成方法通常基于结构-序列(Structure-Sequence，Struct2Seq)框架，但忽略了代码注释的类型信息，例如操作符、字符串等。由于类型信息之间的层次具有依赖性，将类型信息引入已有的Struct2Seq框架并不适用。为了解决上述问题，提出一种基于类型辅助引导的代码注释生成(Code Comment Generation based on Type-assisted Guid-ance，CCG-TG)模型，将源代码视为带有类型信息的n元树。该模型包含一个关联类型编码器和一个限制类型解码器，可以对源代码进行自适应总结。此外，提出一种多级强化学习(Multi-level Reinforcement Learning，MRL)方法来优化所提模型的训练过程。在多个数据集上进行实验，与多种基准模型对比，证明所提CCG-TG模型在所有评价指标上的性能最优。
Code Comment Generation Model Based on Type-assisted Guidance
Code comment generation methods are usually based on the Structure-Sequence(Struct2Seq)framework,but ignore the type information of code comments,such as operators,strings,and so on.Introducing type information into existing Struct2Seq frameworks is not suitable because of the dependency of hierarchy of type information.To solve above problems,a Code Comment Gener-ation based on Type-assisted Guidance(CCG-TG)model is proposed,which treats source code as an n-ary tree with type information.The model includes an associative type encoder and a restricted type decoder,which allows adaptive summary of source codes.In addi-tion,a Multi-level Reinforcement Learning(MRL)method is proposed to optimize training phase of the proposed model.Experiments are conducted on several datasets,the performance of the proposed CCG-TG model is the best on all evaluation metrics by comparing with many benchmark models.
国家科技期刊平台
NSTL
万方数据
