Sensitive Similarity on Programming Assessments Expecting Highly Similar Submissions

Some programming assessments expect highly similar student submissions, adding complexities in detecting plagiarism. There are a number of automated similarity detectors dedicated for such assessments. However, they either are not practical (being integrated with a programming workspace) or might be less effective (relying only on subtle variations). This paper presents a practical approach that exclusively relies on student submissions. It pairwise compares the submissions and sort them based on sensitive similarity, which considers both syntax and superficial similarities. Programs that are similar at syntax level will still be sorted based on their superficial similarity. According to our evaluation involving 610 programs from software design pattern assessments, employing sensitive similarity is more effective than employing the syntax similarity alone. However, it takes longer processing time as it involves more computation. The overall performance is affected by the selection of the similarity algorithm.