Value difference metric for student knowledge level initialization in a learner model-based adaptive e-learning system

Abstract

Web-based learning systems give students the freedom to determine what to study based on each individual learner’s learning goals. These systems support learners in constructing their own knowledge for solving problems at hand. However, in the absence of instructors, learners often need to be supported as they learn in ways that are tailored to suit a specific learner. Adaptive web-based learning systems fit in such situations. In order for an adaptive learning system to be able to provide learning support, it needs to build a model of each individual learner and then to use the attribute values for learner as stored in the model to determining the kind of learning support that is suitable for each learner. Examples of such attributes are learner knowledge level, learning styles and learner errors committed by learners during learning. There are two important issues about the use of learner models. Firstly, how to initialize the attributes in the learner models and secondly, how to update the attribute values of the learner model as learners interact with the learning system. With regard to initialization of learner models, one of the approaches used is to input into a machine learning algorithm attribute values of learners who are already using the system and who are similar (hence called neighbors) to the learner whose model is being initialized. The algorithm will use these values to predict initial values for the attributes of a new learner. Similarity among learners is often expressed as the distance from one learner to another. This distance is often determined using a heterogeneous function of Euclidean and Overlap measures (HOEM). This paper reports the results of an investigation on how HOEM compares to two different variations of Value Difference Metric (VDM) combined with the Euclidean measure (HVDM) using different numbers of neighbors. An adaptive web-based learning system teaching object oriented programming was used. HOEM was found to be more accurate than the two variations of HVDM. Key words: learner modeling, initialization, web-based learning, nearest neighbors, overlap measure, knowledge level, object oriented programming