Code programming language classification method using characterization information of each layer of CodeBert

Abstract

The invention provides a code programming language classification method based on CodeBert, and belongs to the technical field of computer application. According to the technical scheme, the method comprises the following steps that (1) an original data set is processed, noise in the original data set is removed, and word segmentation is conducted through a BPE method; (2) extracting characterization information from each layer of the CodeBert as a characterization information sequence, and paying attention to the layer capable of providing important characterization information by using a bidirectional recurrent neural network (Bi-LSTM) and an attention mechanism; and (3) training the constructed model based on a corpus to obtain a code programming language classification model. The method has the beneficial effects that the programming language type to which the source code belongs can be quickly identified, and the cost of manually classifying the source code programming language is reduced.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a code programming language classification method utilizing CodeBert representation information of each layer. Background technique [0002] In the process of software development and maintenance, programmers often encounter various problems that are difficult to solve. It has become a mainstream solution to publish posts on programmers' question-and-answer websites to seek solutions to problems. But programmers usually use different programming languages ​​(such as Java, Python, C#, JavaScript, etc.) to complete different development tasks. In most cases, the solutions required for problems in different programming languages ​​are also different. Therefore, when a developer publishes a post on a question-and-answer website (such as Stack Overflow, etc.), marking the type of programming language used can help the questioner find the corresponding solution quickly, and the websi...

AI Technical Summary

Problems solved by technology

[0005] Most of the existing work is based on traditional machine learning methods to establish classification models, such as random forest classifiers, naive Bayesian classifiers, etc., but the classification results based on traditional machine learning classification methods are not good

Although a few classification methods based on deep learning models (such as CNN, RNN, etc.) have improved compared with traditional machine learning methods, they still cannot achieve satisfactory performance.

Images