abstract Hedderik van Rijn

Hedderik van Rijn (University of Groningen, the Netherlands)

Personalized, Adaptive Learning based on Cognitive Models Increases Learning Efficiency for Factual Information

It is well known that the schedule of practice partly determines the efficiency of learning sessions, and thus the retention of the learned materials. Even the classical Leitner method for learning factual information is based on this principle, as better encoded items are practiced less often. By calculating the optimal distance between repetitions of a to be learned item, large learning gains can be obtained. Most methods that aim for optimizing the schedule of practice adapt the schedule based on whether a learner provides a correct, or an incorrect answer. However, even when an item is correctly answered, the speed by which an answer is given could be used to assess how well an item is encoded in memory. In this talk, I will present an adaptive learning system that is based on computational cognitive models of the human long-term memory system. This system keeps track of the internal activation of each to-be-learned item, and updates the internal activation after each presentation. Based on this activation value, the system determines which item needs to be practiced at what point in time, or whether the learner is ready for the presentation of new items. We have tested this system in multiple experiments demonstrating learning gains of 10 to 20%, and it is now used by a large Dutch publishing house in their online systems associated with all their secondary education learning materials. I will also discuss recent work with this system that suggests that the internal parameters of the system are better predictors of how well an item is mastered than the score on a test, that these parameters are stable over time and relatively stable over materials, and how these parameters correlate with other, more traditional measures of learning aptitude.