Researchers Track Facial Expressions to Improve Teaching Software
Research from North Carolina State University shows that software which tracks facial expressions can accurately assess the emotions of students engaged in interactive online learning and predict the effectiveness of online tutoring sessions.
“This work is part of a larger effort to develop artificial intelligence software to teach students computer science,” says Dr. Kristy Boyer, an assistant professor of computer science at NC State and co-author of a paper on the work. “The program, JavaTutor, will not only respond to what a student knows, but to each student’s feelings of frustration or engagement. This is important because research shows that student emotion plays an important role in the learning process.”
The researchers used the automated Computer Expression Recognition Toolbox (CERT) program to evaluate facial expressions of 65 college students engaged in one-on-one online tutoring sessions. The researchers found that CERT was able to identify facial movements associated with learning-centered emotions, such as frustration or concentration – and that the automated program’s findings were consistent with expert human assessments more than 85 percent of the time.
The researchers also had the students report how effective they felt the tutorial was, and tested the students before and after each tutoring session to measure how much they learned.
The researchers used observational data from CERT along with student self-assessments and test results to develop models that could predict how effective a tutorial session was, based on what the facial expressions of the students indicated about each student’s feelings of frustration or engagement.
“This work feeds directly into the next stage of JavaTutor system development, which will enable the program to provide cognitive and emotion-based feedback to students,” says Joseph Grafsgaard, a Ph.D. student at NC State and lead author of the paper.
The paper, “Automatically Recognizing Facial Expression: Predicting Engagement and Frustration,” will be presented at the International Conference on Educational Data Mining, being held July 6-9 in Memphis, Tenn. The paper was co-authored by Joseph Wiggins, an undergraduate at NC State; Dr. Eric Wiebe, a professor of science, technology, engineering and math education at NC State; and Dr. James Lester, a professor of computer science at NC State. The research was supported by the National Science Foundation.
Note to Editors: The study abstract follows.
“Automatically Recognizing Facial Expression: Predicting Engagement and Frustration”
Authors: Joseph F. Grafsgaard, Joseph B. Wiggins, Kristy Elizabeth Boyer, Eric N. Wiebe and James C. Lester, North Carolina State University
Presented: July 6-9, International Conference on Educational Data Mining, Memphis, Tenn.
Abstract: Learning involves a rich array of cognitive and affective states. Recognizing and understanding these cognitive and affective dimensions of learning is key to designing informed interventions. Prior research has highlighted the importance of facial expressions in learning-centered affective states, but tracking facial expression poses significant challenges. This paper presents an automated analysis of fine-grained facial movements that occur during computer-mediated tutoring. We use the Computer Expression Recognition Toolbox (CERT) to track fine-grained facial movements consisting of eyebrow raising (inner and outer), brow lowering, eyelid tightening, and mouth dimpling within a naturalistic video corpus of tutorial dialogue (N=65). Within the dataset, upper face movements were found to be predictive of engagement, frustration, and learning, while mouth dimpling was a positive predictor of learning and self-reported performance. These results highlight how both intensity and frequency of facial expressions predict tutoring outcomes. Additionally, this paper presents a novel validation of an automated tracking tool on a naturalistic tutoring dataset, comparing CERT results with manual annotations across a prior video corpus. With the advent of readily available fine-grained facial expression recognition, the developments introduced here represent a next step toward automatically understanding moment-by-moment affective states during learning.