A multilingual automated web usability evaluation agent

The research underlying this thesis explored the development of a customised, intelligent and automated approach to web usability evaluation. An extensive survey of existing web usability evaluation tools was carried out to identify to weaknesses that could be investigated. As result three different issues are addressed in this thesis: Improving and testing usability guidelines particularly for languages other than English; Customising the evaluation; Developing an intelligent (capable of learning) evaluation technique. This thesis presents a new methodology that uses agent technology, which can act and interact on behalf of its owner (the webmaster), to evaluate web pages. The evaluation involves two kinds of customisation, one which reflects the users' tastes and the other the aims of the webmaster. In investigating customisation of web pages to reflect users' tastes the research considered applying this multilingual interface agent approach to the evaluation of multilingual pages in scripts other than the usual Latin. But no guidelines appear to exist for such scripts thus the first difficulty in assessing non-English web pages is the lack of any reliable guidelines. In order to explore multilingual evaluation the researcher first had established guidelines and chose to investigate Arabic. As result usability guidelines for Arabic were established via usability testing. The guidelines are an interesting result of the research in themselves. This thesis presents a set of usability guidelines appropriate for evaluating Arabic web pages produced by testing 196 Arabic users. Also, it validates some of the current usability guidelines for Latin scripts. An interesting variation appeared between the presentations of the two dissimilar scripts, these variations affect font size, emphasized text presentation, the number of links in the web page and the meanings associated with colours. The second form of customisation is represented in the ability to modify the usability evaluation to reflect the webmaster's preferences. This requires an intelligent approach involving learning. Three different kinds of learning were considered; fuzzy average learning, fuzzy learning and Q-learning. All are examined in this thesis in order to identify the most appropriate approach to apply. As the multilingual interface agent learns form its webmaster, Q-learning produced the most accurate evaluation. This thesis represents a useful first step towards multilingual, intelligent, automated web usability evaluation using an agent technique. The automated web usability, multilingual interface agent developed can be customised to suit its users and improve its evaluation in response to the needs of its owner.