Gaze-Driven Adaptive Learning

Overview

For my Master's thesis at RPTU Kaiserslautern (in collaboration with DFKI), I studied how semantic priming and real-time adaptive interventions can improve comprehension and engagement in online learning environments. Using eye-tracking and behavioral measures, I analyzed both learning outcomes and process-level changes to understand how these interventions influence how people read and learn.

The Challenge

Online learning environments often present content as static text, regardless of whether a learner is struggling or breezing through. This one-size-fits-all approach fails to support learners at the moment they need it most. The question: can we use real-time gaze data to detect comprehension difficulty and intervene adaptively without disrupting the reading experience?

Research Approach

• Designed and prototyped adaptive features including AI-generated real-time summaries, interactive word clouds, and dynamic text highlighting
• Planned and conducted controlled user studies to evaluate both usability and cognitive impact of interventions
• Collected eye-tracking data (fixations, saccades, reading speed) alongside comprehension scores and self-report measures
• Analyzed behavioral data using both qualitative and quantitative methods to assess effectiveness and inform design iteration
• Collaborated across interdisciplinary teams to align research insights with interface design and system logic

Why these methods?

Eye-tracking was essential because comprehension difficulty isn't always visible in outcomes alone. Someone can score well on a test but struggle significantly during reading. Gaze data captures that process-level struggle in real time. A controlled study design let us isolate the effect of specific interventions, and self-report gave us the subjective experience layer to complement the behavioral measures.

Key Findings

• Adaptive support fundamentally changed the way users engage with and process learning content, even when comprehension score improvements were modest
• Comprehension improvements depended on factors like language proficiency, suggesting adaptive systems need to account for individual differences
• Process-level measures (reading behavior) revealed intervention effects that traditional outcome measures alone would miss
• Real-time interventions shifted gaze patterns in ways consistent with deeper processing of key content areas

Impact

This research demonstrates that adaptive support doesn't just improve what people learn; it changes how they learn. The findings contribute to the broader field of AI-enhanced learning and show how methods from cognitive science, HCI, and experimental psychology can be applied to real-world UX challenges in digital learning environments.