Towards a Classification for Programming Exercises

When researchers of the AIEDCS (AI-supported Education for Computer Science) community want to exchange programming exercises as baselines for e.g., evaluation purposes, several questions related to the difficulty of exercises will arise: What kind of programming exercises are supported by an intelligent learning environment? How difficult are the programming exercises? In this paper, we investigate programming exercises supported by fifteen existing intelligent learning environments for the domain of programming and have learned that these exercises can be classified into three classes: 1) exercises with one single solution, 2) exercises with different implementation variants, and 3) exercises with different solution strategies. The contribution of this classification is two-fold. First, it can be used to help designers of intelligent learning environments for programming apply/devise appropriate modeling techniques for a specific class of programming exercises that are intended to support the programming/learning process of students. Second, it helps researchers of the AIEDCS community communicate more accurately when they want to discuss programming exercises.