MLTutor: An Application of Machine Learning Algorithms for an Adaptive Web-based Information System

Article type: Research Article

Authors: Smith, A. Serengul Guven | Blandford, Ann

Affiliations: School of Computing Science, Middlesex University, Trent Park, Bramley Road, London, N14 4YZ, UK. serengul1@mdx.ac.uk; http://www.cs.mdx.ac.uk/staffpages/serengul/ | UCL Interaction Centre, University College London, 26 Bedford Way, London, WC1H 0AP, UK. A.Blandford@ucl.ac.uk; http://www.uclic.ucl.ac.uk/annb/

Abstract: One problem that commonly faces hypertext users, particularly in educational situations, is the difficulty of identifying pages of information most relevant to their current goals or interests. In this paper, we discuss the technical feasibility and the utility of applying machine learning algorithms to generate personalised adaptation on the basis of a user's browsing history in hypertext, without additional input from the user. In order to investigate the viability of this approach, we developed a Web-based information system called MLTutor. The design of MLTutor aims to remove the need for pre-defined user profiles and replace them with a dynamic user profile-building scheme in order to provide individual adaptation. In MLTutor, this adaptation is achieved by a combination of conceptual clustering and inductive machine learning algorithms. An evaluation technique that probes the detailed effectiveness of the adaptation is presented. The use of dynamic user profiles has been shown to be technically feasible; however, while a superficial evaluation indicates that it is educationally effective, the more thorough evaluation performed here shows that the positive results may be attributed to other causes. This demonstrates the need for thorough evaluation of adaptive hypertext systems.

Journal: International Journal of Artificial Intelligence in Education, vol. 13, no. 2-4, pp. 235-261, 2003