Effects of Pedagogical Agent-Generated Summaries on Video-Based Learning

Evidence from Eye-Tracking and EEG

Bibliographic Data

ID22044987
AuthorsLei Yuan (0000-0002-0694-8631, Ministry of Education of the People's Republic of China), Jiyuan Xu (0009-0003-1647-8164, Guangxi Normal University), Zehui Zhan (0000-0002-6936-1977, South China Normal University)
Year2025
Volume16
Issue1
Pages39
Publication date2025-12-29
Peer ReviewedYes
Open AccessYes
TypeARTICLE
VenueEducation Sciences (JOURNAL)
Journal identifiersISSN: 2227-7102 • E-ISSN: 2227-7102
PublisherMDPI AG (PUBLISHER • IT)
DOI10.3390/educsci16010039
OpenAlexW7117647337
LanguageEN
Citations received1
References cited39

As an emerging learning support technology, large language model-powered pedagogical agents demonstrate significant potential in enhancing video learning effectiveness, yet the underlying cognitive mechanisms remain inadequately elucidated. This study employed a multimodal approach combining EEG and eye-tracking to investigate the effects of AI-generated mind maps and text summaries on learning performance and cognitive processing. Following data screening, 80 valid datasets from education majors were randomly assigned to three groups: mind map summary (PA-MMS, n = 27), text summary (PA-TS, n = 28), and control (NPA, n = 25). Results showed both experimental groups achieved significantly higher post-test scores than controls, with PA-MMS demonstrating the strongest performance (d = 3.78). EEG evidence indicated pedagogical agents reduced Theta activity (decreased working memory load) while PA-MMS enhanced Alpha activity (superior attention control). Eye-tracking revealed differentiated strategies: PA-MMS exhibited networked fixation patterns facilitating integration; PA-TS demonstrated linear scanning. Delayed testing showed PA-MMS achieved the highest retention (96.8%). Correlations confirmed posttest scores negatively correlated with Theta (r = −0.46) and extraneous load (r = −0.61), positively with germane load (r = 0.54). Mind maps simultaneously reduced extraneous load (d = 1.26) while enhancing germane processing (d = 1.15), representing a shift from static scaffolds to AI-mediated generative support

Cognition · Cognitive Load · Electroencephalography · Generative model · Working memory · EEG and Brain-Computer Interfaces · Gaze Tracking and Assistive Technology · Visual and Cognitive Learning Processes

Tools

Open DOIOpen Access