Clustering Introductory Computer Science Exercises Using Topic Modeling Methods

Manually determining concepts present in a group of questions is challenging and time-consuming process. However, the process an essential step while modeling virtual learning environment since mapping between using mastery level assessment recommendation engines required. In this article, we investigated unsupervised semantic models (known as topic techniques) to assist computer science teachers task propose method transform Computer Science 1 teacher-provided code solutions into representative text documents, including structure information. By applying nonnegative matrix factorization latent Dirichlet allocation techniques, extract underlying relationship validate results external dataset. We consider interpretability learned 14 university professors' data, confirm six semantically coherent clusters current Moreover, topics comprise main test dataset, achieving 0.75 normalized pointwise mutual information metric. The metric correlates with human ratings, making proposed useful providing semantics for large amounts unannotated code.