Code annotation generation method based on machine translation model

Abstract

The invention discloses a code annotation generation method based on a machine translation model. The code annotation generation method comprises the steps that an obtained code corpus containing annotations is divided into a source code method sequence corpus and a corresponding source code annotation corpus; inputting source code annotations in a source code annotation corpus into the seq2seq model; extracting a source code method sequence to generate an abstract syntax tree, replacing identifiers of nodes with node categories, performing structured traversal to obtain a sequence with reserved structure information, and inputting the sequence into a seq2seq model; inputting the structural information into a coding layer to generate a coding output of an implicit state sequence and the structural information, inputting the coding output into an attention mechanism layer, and calculating a context vector according to the matching degree between the coding output of the implicit state sequence and the structural information; inputting the coding output of the structure information and the context vector into a decoding layer, and calculating the sequence probability distribution ofthe training sample target; a sequence-based machine translation model is generated from the source code annotations and the sequence probability distribution.

Description

technical field [0001] The invention relates to the field of software development, in particular to a code comment generation method based on a machine translation model. Background technique [0002] With the continuous development of the Internet era, it involves more and more fields, and the various platform software systems applied to it are becoming larger and larger, which adds complicated difficulties to the maintenance work of later software maintenance personnel. In software engineering, reading code is a basic task, and generally speaking, software maintainers are usually not the original developers, and personnel turnover causes developers to have to take over unfamiliar software code, which also needs to be dealt with in code reviews Lots of unfamiliar source code. Another study shows that in the development of large-scale enterprise applications, developers often spend more time reading code than writing code. At this point, code comments can help in this matt...

AI Technical Summary

Problems solved by technology

[0010] These two ideas are opposite to natural language processing. The generation of code comments uses code fragments to generate comments, and the source code written in a programming language has a clear structure. Existing methods have not utilized rich structural information to improve Model Accuracy

Secondly, since the code contains many variable names, the established dictionary will be very large. For example, there are 790,000 unique identifiers in the data set. If the common training dimension of 30,000 is used, more than 95% of the vocabulary will be regarded as unknown. According to the research shows that it is unreasonable

Images