Investigating gaze interaction usability for web browsing

Abstract

Many websites are built with the basic assumption that conventional input devices, such as keyboards and mice, are to be used as the primary input modalities. Such peripherals require a certain level of dexterity to operate and as a result, people with severe motor limitations, both temporary and permanent, are inherently excluded. Assistive technologies are used to overcome such barriers, however, these come with their own limitations which are further accentuated when interfaces are not designed with accessibility in mind. This exploratory study will focus on gaze interaction for web browsing using low-cost eye trackers in an effort to shed more light on accessible user-agent design for eye-tracking users. In recent years, major advances have been made in this area to facilitate gaze interaction for web browsing, with projects such as GazeTheWeb, providing a native user agent built specifically for eye-trackers. Notwithstanding, certain web-based interaction scenarios are still highly problematic and present usability challenges for gaze interacting users. This work will focus on two specific problem areas generally found on websites, namely: (a) high link density areas (HLDA) and (b) expandable navigation menus. HLDAs presents a challenge due to the level of interaction accuracy required to select links from a cluster of closely placed links, while expandable navigation menus generally assume the use of click, tap or hover interaction patterns, none of which is afforded by an eye-tracker. GazeTheWeb, considered in this thesis to be the gold-standard technology, provides out-of-the-box interaction patterns, and this work will study their usability in these problem scenarios and compare them to two novel approaches developed through an iterative design method. The emerging interaction patterns, namely Quadtree-based Link Selection with Secondary Confirmation (QLSSC) for inline link selection and Hierarchical Re-Rendering of Navigation Menus (HRNM) for expandable navigation menu interaction, were empirically evaluated against the gold-standard approach through a lab-based single-blind study. A mixed-method approach was adopted to be able to study the proposed interaction patterns from both quantitative and qualitative perspectives. Results for the emerging patterns are encouraging, and a discussion on their merits and challenges is also provided. This study also presents Cactus, a purpose-built cross-platform web browser, which was built as an experimental tool to study the emerging interaction patterns, but which also addresses the need for an inclusive web-browser available across platforms. Initial results show that Cactus provides a usable and ergonomic environment for eye-tracker only interaction, even if used with low-cost eye-trackers.