Tell the Algorithm When to Care

Eye tracking technology provides an advanced interface that will benefit a number of industries from marketing to healthcare. In recent years, Apple and Google have acquired companies developing this technology (SMI and Eyefluence, respectively). These movements likely point to a need to optimize in the VR/AR space but eye-tracking can be expanded to number of use cases.

Eye tracking technology allows a developer to track which direction a user is looking and for how long. Eye-tracking is already being used as an interface. For example, the start-up xLabs provides real-time eye tracking with any webcam. Users can look in a particular direction and trigger an interaction with an application. In the case of VR/AR, experiences will be designed to respond to eye-tracking to deepen the sense of immersion.

In the world of machine learning, eye tracking provides a unique opportunity to make our algorithms even better. Barret, et. al., show that human attention as measured by eye-tracking technology can be used to improve a number of standard natural language processing problems.

The authors develop a RNN model that integrates text and eye-tracking measures (i.e. fixation duration time) and show significant improvement in a number of NLP tasks (sentiment analysis, abusive language detection, and grammatical error detection). They train the neural network to learn features from both gaze and text and then can use model to classify the text alone. The reading behavior essentially gives weight to words and tells the algorithm where to focus more.

Barret, et. al., mentions a number of other that have improved accuracy with gaze tracking (here, here, and here). Worth adding to the reading list, it is interesting example of how new technology will be integrated into machine learning process.