Development and validation of weighted biclustering for data mining: IT usability evaluation
Abstract
Usability practitioners need to know how to optimize usability evaluation outcomes. Yet, traditional meta-analysis methods are not applicable to synthesizing the empirical results from usability evaluation studies which is needed in order to optimize outcome measures. Hence a new data mining method, the weighted biclustering, was developed. A new meta-analytic methodology based on the n-corrected effect sizes with fitted asymptotic curve was established and applied to 38 overall discovery rates of usability problems derived from experiments which were reported in publications. The newly developed node-deletion algorithm for weighted biclustering resolves the problems associated with equal weights for data. From the validation with the real data of usability evaluation studies, the node-deletion algorithm for weighted biclustering showed better performance than Cheng and Church (2000)'s node-deletion algorithm, in terms of visual distinction of biclusters, statistical significance of biclusters and usefulness for finding hidden factors. Through the weighted biclustering method and the meta-analytic methodology, the effects of usability evaluation methods, unit of evaluation, expertise of users or evaluators, fidelity of evaluated systems, task type, time constraint, report type, interaction effects and the impact of the number of subjects on problem discovery rates were investigated. The data derived from this study indicates a '9±1 subjects' rule for detecting 80% of usability problems using think aloud and heuristic evaluation methods. However, for cognitive walkthrough method it requires 19 subjects to detect 80% of usability problems.
Degree
Ph.D.
Advisors
Salvendy, Purdue University.
Subject Area
Industrial engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.