Evaluating Control-Flow Graph Similarity for Grading Programming Exercises

Programming has become a fundamental skill in the current digital era. A formal programming course relies on an autograder to score student works. However, the usual black-box method only compares the output instead of adequately examining the code structure. As such, another method is required to measure the structure of the student submission code to give fairer scores. In this paper, an experiment is conducted using a control-flow graph (CFG) comparison algorithm to measure the similarity between student submission code and reference code from the instructor, followed by an analysis of the results obtained from the experiment. The comparison was made using the Hu Algorithm [1], based on previous CFG similarity measurements presented by Chan and Collberg [2]. Through the experiment and analysis process, it is concluded that the CFG comparison method implemented in this research is better applied to boost students who got low scores due to minor mistakes rather than be applied to the entire student submissions as the primary scoring algorithm.