Abstract
Artificial Intelligence–based educational assessment has increasingly become an important area of educational research, particularly in relation to measurement innovation. The development of AI in assessment has expanded beyond automation and efficiency, raising critical issues concerning validity, reliability, fairness, interpretability, human oversight, and alignment with human judgment. This study aims to map and synthesize measurement innovations in Artificial Intelligence–based educational assessment through a hybrid approach that combines a systematic literature review and bibliometric analysis. Specifically, this study identifies research trends, dominant themes, forms of innovation, methodological issues, and future research agendas in AI-based educational measurement. This study used scholarly articles from the Scopus database published between 2021 and 2026. The article selection process followed the stages of identification, screening, eligibility assessment, and inclusion, resulting in 107 relevant articles. Bibliometric analysis was employed to map publication trends, dominant keywords, sources, authors, institutions, countries, and patterns of scholarly collaboration. Meanwhile, the systematic literature review was used to synthesize key themes, innovation models, methodological concerns, and future directions. The findings show that research on AI-based educational assessment developed rapidly during the 2023–2026 period. The major themes identified include automated scoring, automated feedback, multimodal assessment, learning analytics, and human-aligned validation. The literature also indicates that AI-based measurement innovations are not limited to improving assessment efficiency but also emphasize the quality and trustworthiness of assessment processes, particularly in relation to validity, reliability, fairness, interpretability, human oversight, and consistency with human assessment. This study concludes that the development of AI-based educational assessment is shifting from assessment automation toward trustworthy AI-based educational measurement. Such measurement should be valid, transparent, multimodal, and supported by meaningful human involvement. The findings provide a conceptual and bibliometric foundation for researchers, educators, and policymakers in developing AI-based educational assessment systems that are methodologically robust, ethically responsible, and aligned with educational measurement principles.