Reading Progress Tracking Using Convolutional Neural Networks on High-Noise Eye-Tracking Data

Abstract The paper is devoted to studying the methods for tracking the reading progress on eye-tracking data using deep learning neural networks. An architecture of the autoencoder neural network is developed that is intended for efficient use the spatial and temporal information. A data augmentation method is proposed that generates high noise data and preserves information about the correspondence of each gaze fixation to a corresponding word. The quality of the neural network model is experimentally evaluated on noisy data.