Artificial Intelligence in Automated Educational Assessment: A Systematic Review and Integrative Framework

This project presents a systematic, multi-dimensional analysis of Artificial Intelligence (AI) applications in educational assessment. As AI technologies continue to transform teaching and learning environments, their integration into assessment practices has become both widespread and increasingly sophisticated. From early rule-based and Natural Language Processing (NLP) systems to contemporary transformer-based architectures and large language models, AI is redefining how learning outcomes are evaluated, monitored, personalized, and optimized. The review synthesizes evidence from over 100 peer-reviewed studies published across diverse educational contexts and domains. Unlike conventional literature reviews that focus narrowly on technical performance, this study adopts a comprehensive thematic framework that evaluates AI in assessment across seven interconnected dimensions. These dimensions include assessment type (formative, summative, diagnostic, adaptive, and performance-based assessment); AI technology (machine learning models, NLP techniques, transformer architectures such as BERT and RoBERTa, and large language models); AI purpose (automated grading, feedback generation, cognitive-level classification, plagiarism detection, learner modeling, and predictive analytics); assessment method (essay evaluation, multiple-choice analysis, project-based evaluation, peer assessment support, and competency mapping); assessment technique (supervised learning, deep learning, semantic similarity models, vector space representations, attention mechanisms, and prompt-based evaluation); contextual application (K–12 education, higher education, MOOCs, language learning, STEM disciplines, and professional training); and methodological limitations (dataset bias, generalizability issues, explainability challenges, ethical risks, and reproducibility concerns). The temporal scope of the analysis highlights the evolution of AI in assessment—from early BLEU-adapted scoring algorithms and vector space models to advanced transformer-based systems such as BERT-, RoBERTa-powered frameworks, and GPT-4-class large language models. This progression reveals a clear shift from surface-level text similarity evaluation toward deep semantic understanding, contextual reasoning, and generative feedback capabilities. Beyond cataloging technologies, the study critically examines recurring methodological weaknesses, ethical implications, transparency challenges, and gaps in cross-context validation. It identifies emerging trends such as autonomous assessment systems, AI-assisted rubric generation, and adaptive feedback loops. By systematically mapping AI techniques against assessment types, pedagogical goals, and contextual implementations, this review makes several key contributions. It proposes a structured conceptual framework for analyzing AI-driven assessment systems, provides a comparative synthesis of technical and pedagogical approaches, identifies research gaps and future research directions, and offers practical insights for educators, researchers, and policymakers. Ultimately, the project advances the discourse on how AI can move from tool-based automation toward pedagogically aligned, ethical, and intelligent assessment ecosystems.