CopyCatch: A Multi-Algorithm Approach for Detecting Code Plagiarism in Academic Submissions

Academic integrity is a critical aspect of programming-based education, where assignments are widely used for evaluation. While working on this problem, we noticed that detecting plagiarism in source code is not straightforward, as students can easily modify copied code using simple techniques such as renaming variables or changing formatting. This work, includes development of CopyCatch, a digital plagiarism or web plagiarism detection system such combines multiple similarity detection techniques. Instead of relying on a single method, we experimented with approaches such as Jaccard Similarity, TF-IDF based Cosine Similarity, Levenshtein Distance, and statistical divergence measures. Each of these methods captures a different aspect of similarity, which helped us achieve more consistent results.One important observation during our implementation was that supporting modern formats like Jupyter Notebooks adds complexity, as they contain both code and explanatory text. We added preprocessing and OCR-based extraction for non-text formats like PDFs and images to fix this. While we were working on the system, we noticed that using just one similarity measure often led to results that weren't consistent. This made us want to use more than one method to make the system more reliable overall. In general, CopyCatch wants to give teachers a way to use and grow their approach. For future work, planned to explore deep learning–based methods to detect deeper semantic similarities between programs.