Building awareness of programming plagiarism and collusion through similarity feedback generation

Programming plagiarism and collusion can be mitigated by informing students about acceptable practices. However, as that solution is often implemented manually, with a lecturer presenting the information at the beginning of the course, it has three drawbacks: it is labour-intensive, the information is necessarily general (and perhaps irrelevant at the time), and students receive no warning when they might be about to perform a potentially dishonest act. This thesis proposes an approach which automates most of the education process, provides personalised information, and warns students if their work might arouse suspicion. Each time a student's program is submitted, it is compared to the programs of other students. Students whose programs have undue similarity will be notified about this and the possible reasons, generated via a similarity feedback generator (which is also developed in this thesis). Three variants of the approach are developed: short segments, long segments, and gamified long segments. Along with the variants, I also develop a similarity detector that can complement the use of the approach to help identify programming plagiarism and collusion. According to my quasi-experiments, students exposed to any of the three approaches become more aware about programming plagiarism and collusion, including futile program disguises. Evidence also suggests that they are less likely to engage in such misconduct, as both the average program similarity and the number of suspected cases are substantially reduced. This might reflect a positive change in students' behaviour or simply an increase in students' awareness of the issues. Further enhancements to the system, a dedicated gamification feature and a focus on similarities that are less likely to be coincidental, also contribute to the promotion of student awareness of programming plagiarism and collusion.