Automated Methods for Webpage Usability & Accessibility Evaluations

Abstract

Because of the rapid growth of the web and its users, web usability and accessibility become more and more important. This chapter describes two automated methods for evaluating usability/accessibility of webpages. The first method, for usability evaluation, is based on interaction logging and analyses, and the second method, for accessibility evaluation, is based on machine learning. In the following Sections 2 and 3, the two methods are described respectively. Several methods have been proposed and developed for usability evaluation based on user interaction logs. Interaction logs can be recorded by computer logging programs (automated logging) or by human evaluators (observational logging). Analysis methods and tools for the former type of logs are well summarized by Ivory (Ivory & Hearst, 2001; Ivory, 2003), and those for the latter type of logs have also been developed (for example, by Qiang (Qiang et al., 2005)). Our method is for analyzing former type of logs. In the survey by Ivory, analysis methods for automatically captured log files for WIMP (Window, Icon, Menu and Pointing device) applications and web applications are categorized in terms of their approaches including metric-based, pattern-based, task-based and inferential ones. Some of the methods with the task-based approach compare user interaction logs for a test task with desired (expected) interaction sequences for the task and detect inconsistencies between the user logs and the desired sequences (Kishi, 1995; Uehling & Wolf, 1995; Okada & Asahi, 1996; Al-Qaimari & Mcrostie, 1999; Helfrich & Landay, 1999; Zettlemoyer et al., 1999). The inconsistencies are useful cues for finding usability problems: for example, an evaluator can find that users selected some unexpected link on a webpage when another link on the page was expected for the test task and that the link selected by the users may have some usability problem in its design (labeling, layout, etc.). The existing methods require widget-level logs for the comparisons. For example, the method proposed by Okada (Okada & Asahi, 1996) requires interaction logs to include data of widget properties such as widget label, widget type, title of parent window, etc. This requirement degrades independency and completeness of the methods in logging user interactions with systems under evaluation. Section 2 in this chapter describes our method that detects inconsistencies between user logs and desired sequences based on logs of clicked points ((x, y) coordinate values). Coordinate values of clicked points can be easily and fully logged independently of what widgets are clicked on. Several existing methods have also