Leveraging LSTM in the fine-grained analysis of the Incubation Effect in Physics Playground

• Main Authors: Talandron-Felipe, May Marie, Rodrigo, Ma. Mercedes T
• Format: text
• Published: Archīum Ateneo 2019
• Subjects: Computer Sciences
• Online Access: https://archium.ateneo.edu/discs-faculty-pubs/160; https://ir.canterbury.ac.nz/bitstream/handle/10092/17831/ICCE2019-Faiza-cameraready.pdf?sequence=3#page=48
• Institution: Ateneo De Manila University

Incubation Effect (IE) refers to the phenomenon where one gets stuck in a problem-solving activity, decides to take a break, and afterwards revisits the unsolved problem and eventually solves it. While studies on IE were all limited to traditional classroom activities, this research aimed to continue the study of IE in the context of a computer-based learning environment and find features that would predict the incidence of revisiting an unsolved problem and its positive outcome. A prior IE model was developed using a logistic regression but the hand-crafted features used were from aggregated data and do not reflect specific characteristics of students’ actions. Further analysis was conducted in this study and used a deep learning technique which significantly improved the performance of the IE model. In order to interpret the learned features of the neural network, a combination of dimension reduction, visualization technique, and clustering were used. It was found that the coarse-grained features are consistent with the fine-grained features but action level features were also discovered which provided more evidence that there was an improvement on how students tried to solve the problem after incubation