Semantic similarity-based program retrieval:a multi-relational graph perspective

1 Introduction Program retrieval aims to enable the flexible retrieval of program snippets based on a natural language query,significantly accelerating software development productivity.Study shows that over 60%of developers perform program retrieval daily[1].Recent years have witnessed an increasing interest in deep learning-based program retrieval,which aims to construct embedding representations for program snippets and queries.However,the distribution of natural languages and programs is inconsistent,resulting in a semantic gap.The critical challenge is to bridge the semantic gap between the programming language and natural language,and accurately measure their similarity.