基于信息增强的源代码语义嵌入模型  

Semantic Embedding Model of Source Code Based on Information Enhancement

作　　者：张春艳 刘福东 熊其冰 ZHANG Chunyan;LIU Fudong;XIONG Qibing(Information Engineering University,Zhengzhou 450001,China)

机构地区：[1]信息工程大学,河南郑州450001

出　　处：《信息工程大学学报》2024年第6期660-666,共7页Journal of Information Engineering University

基　　金：国家自然科学基金青年科学基金(61802435)。

摘　　要：源代码语义提取是代码智能任务的关键步骤,已有源代码分析工作存在源代码语义提取不充分的弊端。针对该问题,提出一种基于信息增强的源代码语义嵌入模型Aug_Src,信息增强对象为源代码序列和源代码解析的抽象语法树(AST)。相关实验结果表明,基于Aug_Src的代码摘要生成方法比标准序列到序列模型的METEOR指标值提高约11.5%,验证源代码语义嵌入模型的可行性和有效性。Source code sematic extraction is a key step in code intelligence tasks.However,the exist⁃ing code analysis efforts do not extract the semantic information of source code sufficiently.To solve this problem,a new semantic embedding model of source code based on information enhancement is proposed,namely Aug_Src.Information enhancement objects are source code sequences and abstract syntax trees(AST)for source code parsing.The experimental results show that the METEOR metric value of the code summarization generation method based on Aug_Src is 11.5%higher than that of the standard sequence-to-sequence model.,which verifies the feasibility and validity of Aug_Src.

关 键 词：源代码分析 抽象语法树 深度学习 自注意力机制 

分 类 号：TP311.5[自动化与计算机技术—计算机软件与理论]