Code annotation classification method based on neural network model

Abstract

The embodiment of the invention discloses a code annotation classification method based on a neural network model. The method includes: constructing a corpus of annotations, and generating a word vector for representation of each annotation word; classifying the annotations, and manually defining classes; carrying out preprocessing and word embedding processing on the annotations to obtain a 250-dimensional word vector of each word; and entering the 250-dimensional word vector of each word into a classification model to carrying out classification processing to obtain a classification result.In the embodiment of the invention, the context features of the annotations are considered, semantic features of the annotations are also considered, attention is assigned according to word importanceweights, text can be better characterized, and classification is also more accurate. Compared with implementation of other technology, the method is suitable for use in classification of various annotations.

Description

technical field [0001] The invention relates to the field of program understanding and neural network technology, in particular to a code annotation classification method based on a neural network model. Background technique [0002] In recent years, with the development of the software industry, the scale and complexity of software are constantly increasing, and the life cycle of software is also getting longer. A large amount of source code in software systems contains comments, which document the programmer's implementation and help developers understand the code. Code comments play an important role in software maintenance and program understanding. Studies have shown that good-quality code comments can significantly improve the efficiency of developers and maintainers in understanding programs. Therefore, improving the quality of code comments will effectively improve the maintainability of software, and then improve the quality of software. [0003] The current metho...

AI Technical Summary

Problems solved by technology

The existing code annotation classification method based on traditional machine learning only considers the text features of code annotations, but does not consider the semantic information of annotations, etc., and the accuracy of classification will be limited.

Moreover, based on traditional machine learning, such as random forest, the training set is generally relatively small, and it is too sensitive to the noise in the training set, which often leads to a certain degree of overfitting, and the generalization ability of new annotations is not strong.

Images