Cross-cultural analysis of the mechanism of difficulty in emotion recognition for children with autism in AI-driven inclusive education environments

Abstract

This study focuses on AI-driven integrated education classrooms and constructs a cross-cultural analysis framework for the mechanism of emotional recognition difficulties in children with autism. It uniformly encodes multiple modalities such as vision, speech, physiology, and behavior, as well as classroom contexts, and outputs interpretable EMSR chain portraits for identifying the sources of risks in the extraction of cues, pragmatic understanding, and social weighing. The example data shows that the galvanic skin response of the three subjects increased from 2.9–4.5 μS to 3.3–5.1 μS during the task stage, and the total training loss decreased from 1.04 to 0.71. The results indicate that this framework maintains high emotional recognition accuracy while having good cross-cultural transferability, and can provide quantitative evidence for teachers to implement low-burden emotional intervention and individualized support.