Beyond Instructed Tasks: Recognizing In-the-Wild Reading Behaviors in the Classroom Using Eye Tracking
Project Overview
The document explores the role of generative AI in education, particularly through the lens of eye-tracking technology to analyze reading behaviors. It critiques earlier research that utilized instructed reading tasks, which may distort authentic reading patterns, and presents a study that contrasts these with natural reading conditions. By developing a mixed-method framework and an AI classifier, the authors successfully differentiate between reading behaviors, underscoring the necessity of ecological validity in comprehending reading processes. Their findings reveal that a nuanced understanding of reading behaviors can enhance educational assessment and instruction. Additionally, the authors propose a taxonomy of reading behaviors that can be leveraged to inform pedagogical strategies and improve learning outcomes, indicating the transformative potential of combining AI technologies with educational research to foster more effective teaching practices.
Key Applications
Eye-tracking technology for reading behavior recognition
Context: Classroom setting with sixth-grade students
Implementation: Conducted a classroom study using Tobii Pro Spark eye-trackers to collect gaze data during instructed and natural reading tasks.
Outcomes: Developed a lightweight 2D CNN that achieved an F1 score of 0.8 for recognizing reading behaviors, providing insights for educators on how students read.
Challenges: Class imbalance in reading behavior data; difficulty in generalizing findings due to the small sample size and variability in reading tasks.
Implementation Barriers
Data Limitations
The study had a limited sample size of 27 participants, which may affect the generalizability of the findings. Additionally, the focus on a specific age group limits the applicability of the results.
Proposed Solutions: Future research should consider larger and more diverse samples that include participants from different age groups and educational backgrounds to enhance the applicability of the findings.
Ecological Validity
The use of a single PDF document for reading tasks may restrict the variability in text layout and content complexity, which can limit the understanding of how different formats may influence reading behavior.
Proposed Solutions: Future studies should explore a variety of reading materials with diverse layouts, content complexities, and formats.
Lexical Complexity
The study only considered reading material with a single level of lexical complexity, which may impact reading behaviors. This limitation affects the understanding of how varying text difficulty influences engagement and comprehension.
Proposed Solutions: Incorporate materials of varying lexical complexities in future research to understand how text difficulty influences reading behavior.
Project Team
Eduardo Davalos
Researcher
Jorge Alberto Salas
Researcher
Yike Zhang
Researcher
Namrata Srivastava
Researcher
Yashvitha Thatigotla
Researcher
Abbey Gonzales
Researcher
Sara McFadden
Researcher
Sun-Joo Cho
Researcher
Gautam Biswas
Researcher
Amanda Goodwin
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Eduardo Davalos, Jorge Alberto Salas, Yike Zhang, Namrata Srivastava, Yashvitha Thatigotla, Abbey Gonzales, Sara McFadden, Sun-Joo Cho, Gautam Biswas, Amanda Goodwin
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai