Divergent research orientations of affective and emotion concepts in artificial intelligence education through a bibliometric analysis

Abstract This study employs bibliometric analysis to systematically review research trends in AI-driven affective and emotional education published in the Scopus database from 2020 to 2025. Following the PRISMA 2020 guidelines, a dual-track keyword search strategy was adopted, ultimately yielding 401 documents for analysis. A five-dimensional coding framework was established encompassing research purpose, measurement channel, research participants, learning environment, and subject domain, with coding reliability confirmed through a triangulation procedure. Three principal findings emerged. First, keyword co-occurrence analysis revealed that the field has coalesced into four core clusters — technical recognition, psychological experimentation, specialized applications, and instructional practice — signaling that AI affective research has matured into a structurally coherent knowledge ecosystem. Second, longitudinal analysis identified 2023 as a critical inflection point: annual publication volume grew from 26 articles in 2022 to 197 in 2025, with the rise of generative AI driving a research shift from passive sensing and detection toward conversational affective scaffolding. Third, chi-square tests across all five research dimensions yielded statistically significant differences, confirming that "Affective" research is oriented toward educational practice while "Emotion" research is oriented toward technological development. This study proposes the Dual Knowledge Production Model of Affective and Emotion AI Research, clarifying the systematic academic division of labor between the two dominant terms and providing an empirically grounded framework to facilitate cross-disciplinary collaboration.