Evaluating CodeClusters for Effectively Providing Feedback on Code Submissions

Full research paper—Most introductory programming courses rely on the use of automated assessment for grading programming assignments. While such systems save teachers’ time by eliminating manual grading, the submissions may not be reviewed by the teachers at all, losing valuable insight into how students solve the assignments. In this paper, we introduce CodeClusters which provides teachers’ a quick overview of general patterns in code submissions. The main features of the system are full-text search and N-gram -based similarity detection model that can cluster and subset the code by various aspects such as AST similarity or software also has It has also an interface for streamlining the writing of feedback to multiple submissions at once. CodeClusters has been primarily designed to be used jointly with automated assessment systems where automated tests would assess the functionality of the code, and CodeClusters would be used for gaining a higher-level view of programming patterns and for writing feedback to students. CodeClusters was evaluated in a think-aloud study with university lecturers responsible for programming courses at two research-first universities. The lecturers were pleased by the possibility of providing better feedback to students quickly, saw it could improve the quality of their courses over solely automated assessment, and expressed interest in using CodeClusters in their own courses.