Extending educational technologies with user models of cognitive, affective and meta-cognitive student states

点击翻译按钮获取中文摘要

Personalized instruction is key for improving not only students' cognitive expertise related to the target domain (e.g., mathematics) but also their affect (e.g., interest about the domain) and meta-cognitive skills (domain independent abilities needed to learn effectively). Given that computers are becoming ubiquitous in today's classrooms, there is an opportunity to develop educational technologies that maximize student learning through personalized instruction adapted to a given student's unique needs. The prerequisite for realizing this adaptation is that technologies can recognize student states of interest (e.g., what students know about the target domain, how they feel during the learning process, and their learning strategies). This task is accomplished by the user model component. Traditionally, user models have focused on students' cognitive states. However, all three elements (Cognitive, Affective, Meta-cognitive, CAM) impact learning outcomes and so need to be taken into account during the modeling process. A key difficulty in doing so pertains to the fact that there is little if any direct information available to the model on higher level states associated with, for instance, affect or meta-cognition.****The long term goal of my research program is to design and evaluate user models for CAM states in a variety of educational contexts and applications. In the short term, my research agenda entails applying machine learning and expert centric approaches to construct models for CAM states based on data coming from two sources: (1) students' interaction with the target technology and (2) sensing devices (eye tracking, EEG). The models will be designed for four educational contexts that have the greatest potential to make an impact in terms of HQP training, research contributions, and practical implications, including analogical problem solving, collaborative activities, creativity in open-ended environments, and learning from teaching robotic agents. The outcomes will correspond to user models for each of these contexts that are capable of recognizing CAM states most relevant to the corresponding context, as well as general techniques for devising and evaluating models from interaction and sensor data.**