ReClass: An Application of Deep Learning for Real-time Reading Detection Using an Eye Tracking Device
Hoa Thi Nguyen, Jens-Patrick Langstrand, Michael Hildebrandt · Proceedings of the Human Factors and Ergonomics Society Annual Meeting · 2018
We demonstrate the ReClass system, a real time reading detection classifier. ReClass detects if a user is reading text or not solely based on the user’s eye movements, without considering the content of the screen. We examined two machine learning approaches. In the first approach, we pre-processed the data using feature engineering and trained a Linear Support Vector Machine classifier. The second approach used a Deep Learning Neural Network; instead of feature engineering, we allowed the neural network to extract features from the raw data. The second approach outperformed the first by about 25%, achieving 96% accuracy. Our tool illustrates how Deep Learning can be a new innovative method for teaching machines to understand human behaviour. We discuss the potential applications of ReClass for educational assessment, medical diagnosis and training.