Dynamic detection of accessibility smells

Abstract

Automatic detection of accessibility problems is mainly performed by checking for compliance with guidelines on the HTML structure of web pages. While this method can find many problems, it has limitations in detecting difficulties that occur during user interaction. The purpose of this work is to find problematic sequences of interaction events, which we call Accessibility Events. These events occur dynamically as the user interacts with the page and can result in automatic detection of accessibility problems, called Accessibility Smells. We focus on visually impaired users interacting with the web through screen readers. Using previously and recently defined Accessibility Smells, we design Accessibility Events and heuristics to detect them. We describe an empirical study with visually impaired users accessing different pages with known Accessibility Smells. Using a logging tool, we capture Accessibility Events and report on their relationship (or lack thereof) with those smells. For the study, we recruited 8 volunteers, who performed user tests in different websites. During the study, we automatically captured the events on the interfaces and found that out of the 100 events detected during the sessions, 64 resulted in accessibility odors and 19 did not. The remaining 17 were inconclusive, but helped to reformulate the current odor heuristics to analyze potential new ones. The results indicate that it is possible to characterize special patterns of Accessibility Events that may be used to detect potential accessibility issues. While further studies are necessary, our findings provide a base ground for the dynamic detection of accessibility problems in web applications.