Abstract:

Modern science offers various probabilistic and mathematical models for students’ learning style modelling. Many of them use expert knowledge and apply key learning style theories from psycho-cognitive field (for example, David Kolb's model, Honey and Mumford's Learning Styles, Anthony Gregorc's Mind Styles, Visual, Auditory and Kinesthetic (VAK) learners’ model, Felder-Silverman Learning Style Model, etc.). Typically, student’s learning style automatic modelling uses reactive approach, i. e. learning style is modelled using massive amount of historical data about learners’ behavioral activities in the virtual learning environment (VLE). A learning-style model classifies students according to where they fit on a number of scales belonging to the ways they receive and process information [Patricio Garcia, 05]. These scales may represent the divisions of learning styles being used in Cognitive Science, Cognitive Psychology and related fields. But in case an example-based approach is used for learning style modelling, no learning style classification is known in advance. Using example-based approach, the real data about students’ behavioral activities in virtual learning environment are studied for capturing stereotypical learning style patterns, and data labeling is made during machine learning process.

When example-based approach for students’ learning style modelling is applied, issues related to explanation of model results and integration of example-based model with the adaptation components that implement adaptation rules based on the well-known learning style classifications from psycho-cognitive theories must be solved. The following questions should be considered: what learning style (in terms and concepts generally accepted in psycho-cognitive theories) is represented by the cluster defined by important behavioral activities and prototype (i. e. in case the model classifies behavioral activities)? How student’s behavioral activities map to the particular learning style? How a teacher should apply learning style suitability index in case of example-based modelling which classifies behavioral activities, but not representative characteristics of learning styles presented by model from psycho-cognitive field? How should we deal with feature interaction when the prediction of learning style cannot be expressed as the sum of the feature effects as the effect of one feature depends on the value of the other feature (i. e. features are correlated)? May we reuse adaptation rules defined for particular learning style model for adaptation according to example-based learning styles? After investigating transfer learning which uses previously acquired knowledge in new learning or problem-solving situations, authors try to find answers to these questions. In the paper, an approach based on transfer learning techniques is proposed for application of example-based students’ learning style model in virtual learning environments where concepts from the well-known psycho-cognitive models have been used historically.

Keywords:

Transfer learning, learning style modelling, Bayesian network, learning personalization, virtual learning environment.