Programming trajectories analytics in block-based programming language learning

Block-based programing languages (BBPL) provide effective scaffolding for K-12 students to learn computational thinking. However, the output-based assessment in BBPL learning is insufficient as we can not understand how students learn and what mistakes they have had. This study aims to propose a data-driven method that provides insight into students' problem-solving process in a game-based BBPL practice. Based on a large-scale programing dataset generated by 131,770 students in solving a classical maze game with BBPL in Hour of Code, we first conducted statistical analysis to extract the most common mistakes and correction trajectories students had. Furthermore, we proposed a novel program representation method based on tree edit distance of abstract syntax tree to represent students' programing trajectories, then applied a hierarchical agglomerative clustering algorithm to find the hidden patterns behind these trajectories. The experimental results revealed four qualitatively different clusters: quitters, approachers, solvers and knowers. The further statistical analysis indicated the significant difference on the overall performance among different clusters. This work provides not only a new method to represent students' programing trajectories but also an efficient approach to interpret students' final performance from the perspective of programing process.

Automated summary

Block-based programming languages provide effective support for K-12 students to learn computational thinking, but current assessment methods lack insights into students' learning process and mistakes. This study proposes a data-driven method to understand students' problem-solving process in game-based block-based programming practice. By analyzing a large-scale programming dataset, the study identifies common mistakes and correction trajectories. A novel representation method and clustering algorithm are applied to discover hidden patterns and distinguish different clusters of students. The results reveal significant differences in overall performance among the clusters, providing insights into students' programming trajectories and performance.