Code Plagiarism Detection Using Graphic Neural Network Based On Abstract Syntax Tree

Abstract

Code plagiarism is a common issue in education and software development, which is difficult to detect accurately using text-based approaches. Conventional methods such as Term Frequency–Inverse Document Frequency (TF-IDF) and cosine similarity tend to focus only on token similarity, making them less effective in handling structural changes in code. Therefore, this study aims to develop a structure-based code plagiarism detection system using Abstract Syntax Tree (AST) and Graph Neural Network (GNN). The proposed method involves parsing source code into AST, representing it as a graph, and processing it using a GNN model in a pairwise scheme. In addition, a comparison is conducted with a baseline method based on TF-IDF and cosine similarity to evaluate model performance. The dataset used consists of both synthetic and real data, which are divided into training and testing sets. The results show that the GNN model achieves excellent performance with an accuracy of 0.9946, precision of 0.9949, recall of 0.9974, and F1-score of 0.9962, while the baseline method only achieves an accuracy of 0.7392 and a recall of 0.6343. These results indicate that the GNN model is more effective in detecting plagiarism, especially in handling structural code modifications. Therefore, it can be concluded that the structure-based approach using AST and GNN outperforms text-based approaches in code plagiarism detection.