[2C1-GS-12] Human interface, education aid: Estimating states in education (1)

[2C1-GS-12-01] Construction of Neural Network Model for Time-series Estimating Mental States of Each Learner in e-Learning Environment and Attempt for its Interpretability

〇Yoshihisa Furusawa¹, Yoshimasa Tawatsuji², Tatsunori Matsui³ (1. Graduate School of Human Sciences, Waseda University, 2. Global Education Center, Waseda University, 3. Faculty of Human Sciences, Waseda University)

Keywords:Learning Analytics, e-Learning, Interpretable model

E-learning environment provides the opportunity to learn contents asynchronously, and Intelligent Tutoring System technology has been also flourished and adopted to such environment. However, such e-learning support system has also difficulty in that the system should take into consideration the learners ’mental states. In this study, we constructed the system which estimated the learner’ mental states based on learners ’ physiological information with time-series neural network technique. We tried to construct individual model but we have several difficult points, for example these dataset are umbalance, noisy and small size. We use not only the methods in previous research to resolve each point but also proposed methods concidering characteristic of mental state. In addition, we adopted activation maximization and smoothgrad to analyze the system. The results indicated that trained model classifies based on what features of inputs.