<![CDATA[Source code plagiarism identificatio requires a system capable of identifying semantic similarity rather than mere textual resemblance. This study utilized a dataset of 1,000 source code files, which after cleaning resulted in 996 individual code samples collected from GitHub repositories. The dataset included various programming languages (Python, Java, JavaScript, TypeScript, C++), divided into 697 training data, 149 validation data, and 149 testing data. The model employed was CodeBERT, configured with a hidden size of 768, 12 layers, and 12 attention heads. CodeBERT generated vector embeddings for each code sample, which were then projected by a Siamese Network to calculate cosine similarity between code pairs. Testing used a threshold of 0.80 to classify plagiarism. The identification results achieved an accuracy of 96.4%, precision of 95.2%, recall of 97.8%, F1-score of 96.4%, and an error rate of 4.6%. The system produced similarity scores and status labels of “plagiarism detected” or “not detected,” demonstrating the effectiveness of the CodeBERT-based approach for adaptive and intelligent code similarity identificatio.]]>
Copyrights © 2025
