Comparison of eye-tracking data with physiological signals for estimating level of understanding

We propose an e-learning content recommendation system that estimates a learner's level of understanding of a second language sentence. The system analyzes the eye-tracking data of a learner reading a text, and automatically selects the next text based on the estimation. This paper describes the system design and experimentally compares the estimation accuracies of two estimation methods (multiple regression and a neural network) and two kinds of learner-response data (eye-tracking data alone and both eye-tracking data and physiological signals). The neural network achieved higher accuracy than multiple regression, and eye-tracking data alone yielded the same or higher accuracy than the combined eye-tracking and physiological data. The average accuracy rate of the neural network using eye-tracking data was 67.86%.1