Leveraging Learning-to-rank Method In Source Code Retrieval

With the increasing scale of modern software development,the importance of code retrieval in the software development process has also increased.New developers and students can learn through code retrieval tools,and experienced developers can speed up development through them.The main contributions of this paper are as follows:1)An overview of Code Retrieval Research:Through the analysis of the top conference literature published in recent ten years,we point the way how to perform source code retrieval research.2)Sorting learning model with fused text features:This chapter starts from the perspective of textual representation,and combines the relevant background of software engineering field to construct a multi-level semantic relevance representation model,and to study the matching problem between natural language and programming language by constructing domain features.Combined with the learning to rank,the effectiveness of the feature system is evaluated by P@K,NDCG,and MAP.3)Deep match model based on code representation:Based on the deep text matching model,this chapter proposes a source code depth matching model for joint modeling of programming language and natural language,and builds a deep code retrieval system based on this.After relevant experiments,the code retrieval model based on the source code deep matching model can achieve better results and get rid of the dependence on fine annotation data.4)Code retrieval system:In order to verify the validity of the previous two chapters from the practical point of view,combined with the research results of the first two chapters,based on the research of the effectiveness of the retrieval algorithm,the system availability is further explored to construct the code retrieval application system.