Emotional Artificial Intelligence in Education: A Systematic Review and Meta-Analysis
Heng Zhang, Yuhan Liu, Meilin Jiang, Juanjuan Chen, Minhong Wang, Fred Paas
Abstract
Abstract Research on emotional artificial intelligence (emotional AI) in education—AI that detects learners’ emotions and/or provides emotional support and AI that produces affective learning outcomes but does not directly provide emotional support—has grown rapidly, yet quantitative syntheses are scarce. We meta-analyzed empirical studies reporting outcomes in knowledge, skills, affect, perception, and behavior. A meta‑analysis was conducted based on 172 articles on empirical research published between 2000 and 2025; 54 studies reported in 49 articles quantify the effects of emotional AI interventions on cognitive and emotional outcomes. Eight types of AI-based cognitive support and five types of AI-based emotional support were identified. We interpreted effects through control-value theory. Cognitive supports are expected to increase perceived control over tasks, while emotional supports regulate achievement emotions; integrated supports should therefore yield stronger outcomes than emotional-only support. Random-effects models were used; heterogeneity and moderator analyses were conducted. We did not perform formal asymmetry tests for outcomes with k < 20; for affect ( k = 31), Egger’s test indicated asymmetry and model-dependent trim-and-fill results are reported. Integrated cognitive and emotional support showed a large effect on knowledge acquisition ( g = 0.88, 95% CI [0.36, 1.40]), a near-medium effect on affect ( g = 0.42, 95% CI [0.06, 0.78]), and a small effect on perception ( g = 0.18, 95% CI [-0.41, 0.77]). In comparison, cognitive-only support yielded a medium effect on knowledge acquisition ( g = 0.52, 95% CI [0.17, 0.87]), a near-medium effect on skill ( g = 0.44, 95% CI [0.19, 0.68]) and affect ( g = 0.48, 95% CI [ 0.33, 0.63]), and a large effect on perception ( g = 0.85, 95% CI [0.50, 1.20]). Integrated support was found to be more effective for knowledge acquisition than cognitive-only support, but less effective for perception, with similar effect on affect. Evidence for emotional-only support is limited (two studies). Findings have implications for the design of AI-supported educational systems and highlight the need for more rigorous experimental research to isolate the unique contribution of AI-based emotional support.