A Quantification of Students Coding Style Utilizing HMMBased Coding Models for In-Class Source Code Plagiarism Detection

Measuring similarity among source codes produced in programming class, hereinafter called 'in-class' source codes, for grading or detecting plagiarisms is a laborious task. A special similarity measuring method for in-class source codes is needed because: (1) they are often too short to extract enough algorithmic features, and (2) they naturally have strong algorithmic similarity since they are made for the same purpose, and it is difficult to distinguish plagiarism and coincidental similarity in them. The contribution of this paper is to quantify the features based on students' coding style instead of algorithmic features. We approximate a student's coding style which is superficial feature of a source code by a stochastic model, called coding model based on Hidden Markov Model and use it for authentification information of an author.