Deeper Modeling Via Affective Meta-Tutoring

Principal Investigator: 
Co-Investigator: 
Project Overview
Background & Purpose: 

Although research shows computational tutors can be successful in triggering meta-cognition, unfortunately when the meta-tutor is removed, students resume shallow behaviors. In line with current theories, we hypothesize that lasting behavioral shifts in learning strategies require first changing students’ cost-benefit beliefs about shallow vs. deep practices during learning activities, and then breaking their old habits and instilling new ones. Moreover, we propose that these changes are more easily accomplished in a supportive social context. To summarize, the work investigates whether meta-tutoring technology supplemented with the technology of affective learning companions will successfully to foster persistent changes in students’ meta-cognitive behaviors.

Setting: 

The setting for this study will research labs at Arizona State University, as well as summer camps offered every year by the Fulton School of Engineering at this university.

Research Design: 

This is a comparative study designed to generate associative and causal evidence. This project collects original data using observation [personal observation] evidence. We will code the verbal protocol data according to relevant themes (e.g., student motivation, student learning events), quantify the coded data and perform statistical analysis to determine the impact of our intervention. We will also rely on machine learning techniques to analyze the sensor data, focusing on its ability to predict student affect and motivation.

Findings: 

Our first two studies were run during the summer of 2010.  They each had 26 high school students using our teachable agent.  The students "taught" the teachable agent by entering executable system dynamics models for simple situations.   Data were collected from MindReader (a facial image interpreter), pressure mouse, posture sensors, screen recordings and log files.  The data are currently being analyzed.

Publications & Presentations: 

Muldner, K., Burleson, W., van de Sande, B. & VanLehn, K. (2010). An analysis of gaming behaviors in an intelligent tutoring system.  In V. Aleven, J. Kay & J. Mostow (Eds), Intelligent Tutoring Systems: 10th International Conference, ITS 2010 (pp. 224-233). Heidelberg, Germany: Springer.

Muldner,K., Burleson, W., & VanLehn, K. (2010). "Yes!": Using tutor and sensor data to predict moments of delight during instructional activities.   In P. De Bra, A. Kobsa & D. Chin (Eds.) User Modeling, Adaptation and Personalization: 18th International Conference, UMAP 2010 (pp. 159-170) Heidelberg, Germany: Springer.

Other Products: 

The experiment plans to develop new cyberlearning techniques that focus on the role of affect in fostering motivation. These techniques will be centered around computational learning companions and strategies they can employ to trigger both short-term learning and long term meta-cognitive shifts toward deeper learning strategies.