Code clone detection method based on GAT graph neural network model

Abstract

The invention discloses a code clone detection method based on a GAT graph neural network model, and the method comprises the following steps: extracting and generating clone code data of a corresponding definition from a programming competition website and an existing code clone data set according to the definition of a clone code; analyzing the code text to generate an AST abstract syntax tree; adding an artificially defined additional edge on the basis of the AST abstract syntax tree to generate a representation graph; inputting the code representation graph into a GAT network model for training to obtain a graph representation vector; splicing representation vectors of the cloning code pairs and inputting the representation vectors into a binary classification network; and judging an output code cloning prediction result. According to the method, the problem that the code semantic clone detection capability is insufficient in the code clone detection field is solved, the code text is converted into the graph structure representation, clone code information is represented from the semantic and structural level, the internal relation of learning clone codes can be accurately obtained, clone code judgment and prediction are carried out, and the code clone detection accuracy is improved.

Description

technical field [0001] The invention relates to a code clone detection method, in particular to a code clone detection method based on a GAT graph neural network model, and belongs to the technical fields of artificial intelligence natural language processing and graph neural network. Background technique [0002] The problem of code cloning is very common in the software development process, and the method of detecting code cloning is also a very classic problem in the field of software engineering. Generally speaking, there are two reasons for code cloning. The first reason is that developers search for relevant codes through some online question-and-answer websites, such as StackOverflow, etc., and then directly copy and paste the code written by others; the second reason is that Some well-designed code fragments are directly reused. Therefore, it can be seen that the impact of code cloning is two-way, both good and bad. The good effect is that excellent code components...

AI Technical Summary

Problems solved by technology

Although GCN can effectively process the first-order neighbors of nodes in the graph and avoid complex matrix operations, the model relies heavily on the structure of the graph

Therefore, models trained on specific graph structures often cannot be directly used on other graph structures

Images