Enhancing Computer Science Education by Automated Analysis of Students’ Code Submissions

Abstract

Lecturers of introductory programming courses are often faced with the challenge of supervising a large number of students. Reviewing a large number of programming exercises is time-consuming, and an automated overview of the available solution approaches would be helpful. In this paper, we focus on source code similarity at the level of students’ selected solution approaches. We propose a method to compare Java classes using variable usage paths (VUPs) extracted from modified abstract syntax trees (ASTs). The proposed approach involves matching semantically equivalent functions and attributes between classes by comparing their VUPs. We define a $F_1$-based similarity measure on how well one student submission matches another. We evaluate our approach using students’ submissions from an introductory programming exercise and the results indicate the effectiveness of our method in identifying different solution approaches. The proposed approach outperforms simplified comparisons and the widely used plagiarism detection tool JPlag in accurately grouping submissions by solution approach similarity.