AI-Based Emotion Recognition in Education: Identifying Dyslexic and Non-Dyslexic Learners

This is the authors' abstract. We don't add key points for this paper.

Facial Expression Recognition (FER) is becoming increasingly relevant in educational contexts, where detecting learners' emotional states can help identify cognitive difficulties, predict engagement, and inform adaptive strategies. This study explores the use of FER to identify students with Specific Learning Disorders (SLDs) based on their emotional responses during the viewing of educational video content. A group of 20 university students, equally distributed in individuals with certified SLDs and a control group, was asked to watch a 6-minute and 34-second explanatory video while their facial expressions were recorded via a webcam. A computer vision model based on the ResNet18 architecture was trained to classify four cognitive-affective states relevant to learning: Boredom, Confusion, Engagement, and Frustration. Emotion data were analyzed both as percentage distributions and as probabilistic time series. Group-level differences were assessed using t-tests and will be further explored using Statistical Parametric Mapping (SPM). Preliminary results indicate distinct emotional patterns between groups, supporting the feasibility of integrating emotion-aware AI into adaptive storytelling systems such as OLOS® for inclusive educational and cultural experiences.