Student Programs Performance Scoring Based on Probabilistic Latent Semantic Analysis and Multi-granularity Feature Fusion for MOOC

Abstract

In order to solve the problem of the low accuracy of automatic scoring for programming questions on MOOC platform, this paper proposed a multi-granularity feature fusion automatic scoring method based on potential semantic analysis. Abstract syntax tree (AST) is used to extract the features of student evaluation programs and standard answer template program, and calculate the similarity of features. According to whether the program is compiled or not, the similarity of multi-granularity features is analyzed by different strategies to score automatically. The experimental results show that the average accuracy of the method proposed in this paper outperforms the dynamic test method and the traditional static method using the test case results only, and the automatic machine scoring results are highly consistent with the human score.