An Experimental Comparison of Three Code Similarity Tools on Over 1,000 Student Projects

Massive courses are gaining popularity. They offer many advantages for the market but predominantly for learners and lecturers. Along with benefits come disadvantages, and in the case of massive programming courses, these are mainly problems with code originality. The following article focuses on three popular tools for code similarity detection: JPlag, Sherlock, and MOSS. In the experiment, we used two samples with student codes containing 530 programs in C and 540 programs in Java. The results revealed the reliability of the JPlag tool compared to others, having a problem with false positives. On the other hand, Sherlock was more reliable in the case of structural code changes. In this context, we also analyzed the most common ways to bypass originality checks.