Sharing user models between interactionally-diverse adaptive educational systems
Rikki Prince · ePrints Soton (University of Southampton) · 2017
Adaptive Educational Systems (AES) will become increasingly important in teaching and learning environments over the next decade, as students demand more personalised experiences. These systems reveal, hide, modify and recommend content that is most appropriate for the current user. To do this they rely on an accurate model of the student, their knowledge, experience and goals. With a growing variety of developers of these systems there are more situations where an experienced student will approach a new adaptive system, and it will not have any user model data with which to adapt; this is known as the cold-start problem. An answer to this is shared user modelling, where data about the student is communicated between adaptive applications. This task becomes more complicated when the applications measure the user in very different ways and therefore have different models to represent the user. This thesis proposes the design of an intermediary user model system that uses authored rules to map between the user model attributes used by different applications to measure the user. A prototype implementation of this theoretical framework is presented here, called the Interactionally-Diverse Intermediary User Modelling System, or IDIUMS. Two evaluations of IDIUMS were performed: a simulation and a user trial. The simulation demonstrated that the rule mapping functions as expected, producing user models that are still representative of the user, in relation to all other user models. The user trial showed that use of IDIUMS did not result in the adaptive applications presenting content at a more appropriate level, as perceived by the user. In determining why the user trial did not demonstrate appropriate adaptations, a review of evaluation methodologies in the AES community was undertaken. This showed that the method implemented for the user trial was in the second most common category of sources of evaluation data, behind expert-measured evaluations like pre-post test.