Artificial Intelligence-Driven Adaptive Assessment Systems: A Systematic Survey

Abstract

Modern educational assessment faces critical challenges in capturing fine-grained student proficiency through static, fixed-form testing methods that impose uniform burdens and overlook dynamic learning trajectories. To address these limitations, artificial intelligence has emerged as a transformative paradigm, shifting traditional static testing toward intelligent, real-time adaptive ecosystems capable of continuous student modeling and personalized evaluation. This paper presents a systematic survey of artificial intelligence-driven adaptive assessment systems, synthesized through a rigorous PRISMA-based review methodology. The study systematically examines both foundational psychometric milestones and recent technological advances across peer-reviewed literature. A novel end-to-end taxonomy is introduced, categorizing the literature into five core pillars: measurement models spanning classical test theory to neural cognitive diagnosis, adaptive item selection algorithms leveraging psychometric information and reinforcement learning, automated item generation driven by transformer models and large language models, automated scoring mechanisms utilizing contextual embeddings, and underlying system architectures paired with learning analytics. Furthermore, this survey provides a comparative analysis of existing methods, identifies critical open research challenges regarding algorithmic bias and model interpretability, and outlines a comprehensive future research roadmap. Ultimately, this work highlights the critical significance of developing trustworthy, explainable, and intelligent adaptive assessment systems for modern education.