Code annotation generation method based on multiple source code representation and recurrent neural network

Abstract

The invention provides a code annotation generation method based on multiple source code representation and a recurrent neural network, comprising the following steps: S1, collecting Java code annotation pairs, and constructing a corpus; s2, in a serialization processing layer, converting the source code in the corpus into a token sequence, an SBT sequence and an API sequence; s3, in an encoder layer, a bidirectional GRU is used as an encoder, a codeseq encoder, an SBTseq encoder and an APIseq encoder are constructed for the three sequences respectively, and information of different levels of source codes is learned; s4, in a decoder layer, using a one-way GRU to construct a decoder, and using a teach forming strategy to train a model; and S5, adding an attention layer behind each encoder in the three encoders, linking attention matrixes input by the three encoders and input by the decoder, learning how to combine each input code by using a full connection layer, and finally outputting code annotations. The code annotation generation accuracy is improved, the efficiency of software developers in the software development process is improved, and the development time is saved.

Description

technical field [0001] The invention belongs to the technical field of code annotation generation, and in particular relates to a code annotation generation method based on multiple source code representations and a cyclic neural network, which is mainly used for generating corresponding annotations for source code fragments input by users. Background technique [0002] In the field of software engineering, when faced with a large number of large-scale software and complex systems, software personnel need to read the code quickly and accurately, and efficiently complete tasks such as software change and maintenance. Code comments can help developers and maintainers to read the source code faster, better understand the design ideas behind the code and the behavior of the code, greatly save the precious time of developers and maintainers, and improve work efficiency. Given this, code comments play a very important role throughout the software declaration cycle. [0003] Writi...

AI Technical Summary

Problems solved by technology

[0003] Manually writing comments is expensive and inefficient, so people try to use computers to automatically generate comments for code

Images