Adapting the scheduling of illustrations and graphs to learners in conceptual physics tutoring

[摘要] This research investigates how to schedule multiple graphical representations in a dialogue-based conceptual physics tutor.Research on multiple graphical representations in tutoring suggests either frequently switching representations or fading from concrete to abstract representations.However, other research communities suggest that the best representation or scheduling can be dependent on various student and tutoring context factors.This thesis investigates whether these factors are important when considering a schedule of representations.Three major hypotheses are investigated.H1: that the best representational format for physics concepts is related to properties of the student and the tutoring context.H2: that it is possible to build models that predict the best representational format using student and tutoring context information.H3: that picking the representational format based upon student and tutoring context information will produce better learning gains than not considering student and tutoring context information.Additionally, this work addresses the question of whether multiple representations produce greater learning gains than a single representation (H4).A first experiment was performed to both investigate H1 and to collect data for H2.ANOVAs showed significant interaction effects in learning between low and high pretesters and between high and low spatial reasoning ability subjects, supporting the first hypothesis.Using the data collected and features describing student and tutoring context information, models were learned to predict when to show illustrations or graphs.That these models could be learned, produce meaningful rules, and outperformed a baseline supports H2.A new modeling algorithm was developed to learn these models by augmenting multiple linear regression to consider certain syntactic constraints.A third study was run to test H3 and H4 and to extrinsically evaluate the adaptive policy learned.One third of subjects had an adaptive scheduling of representations, one third a fixed alternating scheduling, and one third saw only one representation.In support of H3, subjects with high incoming knowledge sometimes perform better when receiving adaptive scheduling over an alternating scheduling, but there are also counter examples.For H4, it is not supported in general: showing only illustrations is best overall, but in some cases some subjects benefit from multiple representations.

[发布机构] the University of Pittsburgh