Davis, Paul (2010-12). Learning Usability Assessment Models for Web Sites. Doctoral Dissertation. Thesis uri icon

abstract

  • This research explores an approach to learning types of usability concerns considered useful for the management of Web sites and to identifying usability concerns based on these learned models. By having one or more Web site managers rate a subset of pages in a site based on a number of usability criteria, the approach builds models that determine what automatically measurable characteristics are correlated to issues identified. To test this, the approach collected usability assessments from twelve students pursuing advanced degrees in the area of computer-human interaction. These students were divided into two groups and given different scenarios of use of a Web site. They assessed the usability of Web pages from the site, and their data was divided into a training set, used to find models, and a prediction set, used to evaluate the relative quality of models. Results show that the learned models predicted remaining data for one scenario in more categories of usability than did the single model found under the alternate scenario. Results also show how systems may prioritize usability problems for Web site managers by probability of occurrence under context rather than by merely listing pages that break specific rules, as provided by some current tools.
  • This research explores an approach to learning types of usability concerns
    considered useful for the management of Web sites and to identifying usability concerns
    based on these learned models. By having one or more Web site managers rate a subset
    of pages in a site based on a number of usability criteria, the approach builds models that
    determine what automatically measurable characteristics are correlated to issues
    identified. To test this, the approach collected usability assessments from twelve
    students pursuing advanced degrees in the area of computer-human interaction. These
    students were divided into two groups and given different scenarios of use of a Web site.
    They assessed the usability of Web pages from the site, and their data was divided into a
    training set, used to find models, and a prediction set, used to evaluate the relative
    quality of models. Results show that the learned models predicted remaining data for
    one scenario in more categories of usability than did the single model found under the alternate scenario. Results also show how systems may prioritize usability problems for
    Web site managers by probability of occurrence under context rather than by merely
    listing pages that break specific rules, as provided by some current tools.

publication date

  • December 2010