A Model for Identifying Frequent Errors in Incorrect Solutions

Identifying frequent errors in student solutions is one of the most important but also most laborious tasks in education, regardless of the subject matter. However, when the topic is computer programming, dealing with errors in program codes is a particularly challenging task for both students and instructors. In this paper, we propose a model for identifying frequent error patterns in solution codes in which source code pairs (wrong-corrected) from an e-learning system are selected and converted into token sequences. Then, differences between the paired codes are clustered to discover the most frequent errors by using the K-mean clustering algorithm. To test our model, we conducted an experiment involving large-scale solution codes collected from the Aizu online judge (AOJ) system in which more than five million solutions have been accumulated. Our experimental results show that frequent error patterns can provide useful hints for both learners and instructors, which can then be used to resolve errors in solutions quickly.