LEGenT: Localizing Errors and Generating Testcases for CS1

Due to large class sizes, providing frequent feedback to CS1 students for programming assignments is a demanding task. We present LEGenT, an automated tool to provide personalized feedback in the form of testcases for student submissions. LEGenT uses a few off-the-shelf tools to divide the submissions into structurally similar and semantically equivalent clusters. LEGenT then generates a testcase for the incorrect submissions in a cluster using a structurally similar correct submission such that the generated testcase exposes the deviation from the correct behaviour. We have evaluated our technique on 26082 real student submissions(2764 correct + 23318 incorrect) across 11 different programming problems. LEGenT localized erroneous statements and generated testcases for 7696 incorrect submissions. We have also presented a novel validation method to prove the accuracy of our results identification of the cause of deviation). We are able to validate 77.46% of LEGenT?s results using our novel validation process.