A model with iterative trials for correcting logic errors in source code

Abstract

It is difficult for students and teachers to detect and correct logic errors in source code. Compilers and integrated development environments (IDEs) have the ability to detect and correct syntax errors but it is also difficult for them to detect and correct logic errors. Although many machine learning approaches have been proposed that can show correction candidates for logic errors, they do not provide guidance concerning how the user should fix them. In this paper, we propose a model for correcting logic errors in a given source code. The proposed model realizes debugging of multiple logic errors in the source code by iterative trials of identifying the errors, correcting the errors, and testing the source code. In this model, in the first stage, a list of correction candidates is provided by a deep learning model, and then the list is given to an editing operation predictor that predicts the editing operation for the correction candidate. To learn the internal parameters of the proposed model, we use a set of solution codes created to solve the corresponding programming tasks in a real e-learning system. To verify the usefulness of the proposed model, we apply it to 32 programming tasks. Experimental results show that the correction accuracy is, on average, 58.64% higher than that of the conventional model without iterative trials.