Intelligent software agents for teaching across the WWW

  • Jorge Bergasa-Suso

Student thesis: Doctoral Thesis


This dissertation describes the creation of new Web-based Teaching (WBT) systems to assist in the use of the Internet, as well as the creation of new intelligent agent systems to monitor user behaviour while browsing the World Wide Web (WWW). A key contribution to knowledge is the creation of a method to infer user learning style from user behaviour while browsing the WWW and the inference riles resulting from the application of this method.
Existing commercial WBT systems provided useful tools to facilitate the use of the Internet. However, most of these systems were designed for distance learning, and not for using the Internet within classrooms, so students couls lose concentration and navigate to unrelated Web sites. Existing commercial WBT systems did not provide intelligent advice on potential sites, consider student activity or provide content-specific filtering of Web pages.
A system called CITA was designed to overcome these limitations. A prototype was created using a standard proxy server as a platform fro testing the effectiveness of filtering methods. The knowledge gained from testing the prototype suggested a need for another type of software tool that provided structured, focused and controlled access to the Internet in an intuitive and non-intrusive way, relying on a minimal network infrastructure. A novel set of tools called iLessons was created to achieve these goals. iLessons enabled teachers to: gather resources from the Internet; and load lessons into student computers. iLessons also provided students with tools to create resource collections and to create coursework. Users considered iLessons to be intuitive and easy to use because it was embedded into a standard Web browser.
The research moved on to create a model of a new collaborative agent system vthat filtered and recommended Web pages to students based on three different dimensions: page relevance; student learning style; and student activity. In order to automatically determine the learning style of students and recommend suitable Web pages, patterns were sought in the way students interacted with a standard Web browser and in the strucure of Web pages that were preferred by each learning style group. Two new intelligent agent systems were createdto record user activity and Web page structure while using Web browsers: Solstice and BUCAgent. Solstice was a first prototype created to test the methodology. BUCAgent was then created to record UI activity information and Web page structure features. The same technology as iLessons was used so that they could be fully integrated with it.
BUCAgent was utilised in a controlled environment while volunteers completed a research task. Collected data was analysed using a data mining engine to find rules and to predict user dimensions of learning style. Rules to predict the Active/Reflective, Sensing/Intuitive and Visual/Verbal dimensions of learning style were found. It also proved that parameters in the way that users interacted with the Internet could be measured to classify users in a number of behavioural groups, such as different learning style models or larger scale psychologica; models. Systems could then adapt their behaviour to suit the behavioural traits of the user.
Date of AwardDec 2005
Original languageEnglish
SupervisorDavid Sanders (Supervisor), Giles Tewkesbury (Supervisor) & Jasper Graham-Jones (Supervisor)

Cite this