Search

In this paper, an intelligent training model is designed for the OJ (Online Judge) system. In the traditional OJ system, students practice on a problem-by-problem basis, and get a score for completing a problem. In the intelligent training mode, all the problems are divided into categories according to knowledge points, and students practice on a category-by-category basis, and get a score for completing a category. Each student has to complete a different number of problems in the same category (the system will intelligently determine whether the student has met the passing conditions), thus achieving the goal of personalized training in a hierarchical manner. In order to reduce the teachers’ workload and improve the classification accuracy, the similarity matrix is constructed based on the similarity of the abstract syntax tree (AST) of the problems’ source code, and the problems are clustered by the K-Means algorithm, and good results are obtained.

We proposed a plagiarism detection approach based on code similarity and student behavior characteristics in educational scenarios. The traditional plagiarism check is based on the code only, which enables that students can escape inspection by modifying a small amount of code. We proposed that if the behavioral characteristics of students when submitting code can be considered, the suspected plagiarism can be more accurately identified. We proposed the concept of code similarity concentration (SCD) with reference to the Gini coefficient idea. SCD can reflect the similarity distribution between all the codes submitted by a student and others' codes. A large value of SCD means that a student's codes are always the most similar to the codes of some particular classmates. In addition, we also extracted other features to help detection. Finally, we classify the plagiarism detection problem as a binary classification problem and use LightGBM to make decisions. The experimental results show that the accuracy is close to 99% and f1-score is close to 98%.