On the Potential of Evolved Parsons Puzzles to Contribute to Concept Inventories in Computer Programming
A.T.M. Golam Bari, Alessio Gaspar, R. Paul Wiegand, Dmytro Vitel, Kok Cheng Tan, Stephen Kozakoff · 2020
Abstract Our goal is to investigate whether techniques to automatically generate practice problems have also potential to assist in constructing Concept Inventories (CI) for computer programming. More specifically, we focus on a specific type of practice problem, Parsons puzzles, aimed at novice programmers. In this study, we propose EvoParsons - an automated way of evolving Parsons puzzle for newbie computer programmers, and more importantly we establish that EvoParsons can be a stepping stone of automating the process of building CI. EvoParsons is a software tool to improve students’ learning of computer programming. It is developed, maintained and distributed by our team. The state-of-the art techniques of building CI largely depends on several iterations of settings among faculties, interviews and surveys from students. This so called Delphi method largely depends on knowledge of domain experts, feedback from students, surveys etc. EvoParsons goal is to automate this process by applying Competitive Coevolutionary Algorithm (CCoEA) and Interactive Evolutionary Algorithm (IEA). In this paper, we first describe EvoParsons working mechanism, its benefits over other existing systems of generating Parsons puzzle. Second, we use EvoParsons’ interaction data with actual student to describe its potential to contribute for building CI. To do so, we perform data driven analysis of EvoParson’s misconceptions that are found at the last generational puzzle of the experiment. We also analyze its interaction log to investigate pedagogical importance of misconceptions in successive generations. Experimental analysis shows that EvoParsons evolves interesting misconceptions, discards trivial ones, maintains an order of misconceptions in its subsequent generations of evolution.