Automatic Assistance to Cognitive Disabled Web Users via Reinforcement Learning on the Browser

Abstract

This paper introduces a proof of concept software reasoner that aims to detect whether an individual user is in need of cognitive assistance during a typical Web browsing session. The implemented reasoner is part of the Easy Reading browser extension for Firefox. It aims to infer the user’s current cognitive state by collecting and analyzing user’s physiological data in real time, such as eye tracking, heart beat rate and variability, and blink rate. In addition, when the reasoner determines that the user is in need of help it automatically triggers a support tool appropriate for the individual user and Web content being consumed. By framing the problem as a Markov Decision Process, typical policy control methods found in the Reinforcement Learning literature, such as Q-learning, can be employed to tackle the learning problem.