Modeling Curiosity in Virtual Companions to Improve Human Learners' Learning Experience

Abstract

A key design aspect for virtual learning companions is their believability. A lot of attention has been paid to emotion modeling which is at the core of believability. However, most of the existing emotion models neglect the epistemology-based emotions, which are knowledge-related emotions that affect the human learning process. Studies have shown that curiosity is an important epistemology-based emotion that positively influences social learning. Hence, modeling curiosity in learning companions may improve human learners' learning experience in a virtual environment. However, existing curiosity models assume simplified cognitive processes and fail to capture multiple sources of curiosity stimuli. In this paper, we propose a novel model of curiosity for learning companions to capture salient curiosity stimuli through a psychologically inspired approach. Our model is built based on Berlyne's theory and considers three most salient appraisal variables in a virtual learning environment, including novelty, surprise, and uncertainty. The model is built on planbased knowledge representations augmented with planning. Two internal processes are modeled for learning companions to demonstrate curiosity: curiosity appraisal and learning. The proposed model of curiosity is implemented in a learning companion and evaluated through user studies. The evaluation results show that the learning companion's curiosity significantly improves human learners' learning experience from multiple aspects.