Repeated Pull Request detection method based on graph neural network

Abstract

The invention discloses a repeated Pull Request detection method based on a graph neural network, and the method comprises the following steps: 1) processing a data set and obtaining data, and obtaining the title, description, commit information, change file information and change code information of a Pull Request through the calling of a GitHub API according to the number of the Pull Request in the data set; and filtering the Pull Request when the change file exceeds 50 or the change code is added and deleted by more than 10000 lines. 2) calculating the similarity of title, description and commit information by using a natural language processing method in combination with cosine similarity; calculating the path similarity of the changed file by using a longest common sub-path algorithm; and calculating the similarity of the positions of the change codes, calculating the length of the overlapped change positions by calculating the specific positions of the change codes in the two PullRequest, and calculating the similarity of the change codes by dividing the total length of the overlapped positions. And 3) based on a large code clone data warehouse, training a graph neural network model by adding a flow abstract syntax tree, a graph matching network and mean square error loss, and calculating the similarity of changed codes. And 4) on the basis of the obtained title similarity, description similarity, commit information similarity, change file path similarity, change code position similarity and change code similarity as feature values and corresponding labels for determining whether to be repeated or not, training a repeated Pull Request detection model by utilizing an AdaBoost algorithm in machine learning.

Description

technical field [0001] The invention belongs to the field of software engineering development and maintenance, and in particular relates to a repeated Pull Request detection method based on a graph neural network. Background technique [0002] In recent years, with the popularity of collaborative development, more and more developers contribute to open source projects by submitting Pull Requests. A typical Pull Request on GitHub includes the following steps: first, contributors follow some well-known developers to participate in attractive projects; then clone the project to the local, add new features or fix bugs, etc.; then contributors initiate Pull Requests to the original repository; then the reviewers of the original repository review the PullRequest, discuss whether the Pull Request meets the requirements, whether the quality needs to be further improved, etc., and give some suggestions; next, the contributors improve according to the reviewer's suggestions And updat...

AI Technical Summary

Problems solved by technology

The work is mainly divided into two categories. One is to calculate the similarity based on the title and description of the Pull request, but there are fewer features considered. There are more features to be considered for the Pull request duplicate detection.

The second category is to obtain similarity through title, description, change file, and change code, and use machine learning to train the model. Although the characteristics are considered adequately, the detection method for the similarity of the changed code is relatively backward, and the code cannot be detected well. Semantic similarity of

Images