PROGRAMMING STYLE ON SOURCE CODE PLAGIARISM AND COLLUSION DETECTION

This paper utilises programming style on a source code plagiarism and collusion detection to both capture obvious attempts of such academic dishonesty (which characteristics are ignored on most detection techniques) and prioritise non-coincidental similarity to the coincidental one (as only the former can raise suspicion). The technique relies on pairwise programming style similarity to deal with the former and dishonesty probability (how significant is the programming style change between the author’s current submission and previous submissions) to deal with the latter. According to our evaluation, programming style similarity can increase precision since when a code is copied, the programming style can be unconsciously shared (especially for novice students). Dishonesty probability increases not only precision but also recall, f-score, and the resulted similarity degree of suspected pairs; the copied code commonly has different programming style in comparison with the student’s usual style (captured from previous submissions). Our detection technique is comparable to a common technique in academia except that it takes longer processing time as more hints are generated and considered.