Recommendation of Others’ Code Based on Behavioral Similarity to Promote Code Refinement Activities in Programming Learning

In recent years, as the importance of programming education has grown, conventional lectures emphasize theability to write code that meets requirements, while guidance on refining code behavior (runtime execution quality)remains implicit. The authors have previously proposed a learning support framework that encourages learners torefine their code behavior through trial and error, implementing a “crop harvesting game” using virtual robots as acase study. While the system included a feature allowing learners to view higher-ranked others’ code based on scores,some learners struggled to effectively utilize the recommended code. To address this issue, this study proposes a coderecommendation method based on behavior similarity. Specifically, it calculates an F-score based on precision andrecall from the robot’s action sequences during code execution, evaluating this as behavior similarity. Its effectivenesswas verified through two stages: an evaluation experiment with three experts and an evaluation experiment with 10university students. Results from the expert evaluation suggested that the proposed method is educationally beneficialcompared to conventional methods. In the student evaluation, particularly in tasks requiring obstacle avoidance, thehigh-similarity condition showed significant score improvements and a tendency toward increased incorporation ofothers’ code, though further verification with larger samples is needed.