Quantifying Novice Behavior, Experience, and Mental Effort in Code Puzzle Pathways

Code puzzles are an increasingly popular approach to introducing programming to young learners. Today, code puzzles are predominantly introduced through static puzzle sequences with increasing difficulty. However, adaptive systems in other domains have improved learning efficiency. This paper takes a step towards developing adaptive code puzzle systems based on controlling learners' cognitive load. We conducted a study comparing static code puzzle pathways and adaptive pathways that predict cognitive load on future puzzles. While the trialled adaptive recommendation policy did not result in better learning, our findings point us towards a different policy which may have a greater effect on learner experience. In addition, we identify predictors of student dropout, and use our experimental data to quantify learners' puzzle-solving experiences into 7 principal component properties and use these factors to suggest approaches for future adaptive systems.