Explainable Artificial Intelligence in Automated Essay Scoring: A Systematic Review of Transparency In AI-Based Educational Assessment

Abstract

Background and Purpose: Artificial intelligence is increasingly used in educational assessment through Automated Essay Scoring (AES), where machine learning and natural language processing models evaluate grammar, vocabulary, coherence, and semantic relevance. Although these systems improve scalability and reduce marking time, increasingly complex models can operate as black boxes, creating concerns about transparency, interpretability, fairness, and accountability. This review examines explainable artificial intelligence (XAI) in AES and summarizes the models, explanation techniques, and implications for transparent educational assessment.

Methods: A systematic literature review was conducted using Scopus, Web of Science, IEEE Xplore, ERIC, and Google Scholar. The historical search window covered literature from 2010 through November 2021. Peer-reviewed studies addressing machine learning or deep learning for essay evaluation together with explainability or interpretability mechanisms were considered. Eligible studies were synthesized qualitatively in accordance with PRISMA 2020 principles.

Findings: Thirteen studies published between 2016 and 2020 met the historical eligibility cutoff. Neural-network-based AES approaches were prominent, while explainability methods included feature importance analysis, attention or saliency visualization, LIME, SHAP, and interpretable model frameworks. Across the literature, explainability was consistently linked to improved transparency, educator understanding, and the ability to scrutinize automated scoring decisions. Theoretical Contributions: The review consolidates research spanning educational technology, natural language processing, and explainable AI and positions interpretability as a core requirement for responsible automated assessment. It also organizes the principal AI model families and explanation strategies used in AES before December 2021.

Conclusions and Policy Implications: Explainable AES offers a pathway toward more transparent and trustworthy AI-assisted assessment, but automated scores should remain subject to human oversight and institutional validation. Educational institutions should prioritize auditable models, clear explanation mechanisms, bias monitoring, and governance procedures before deploying automated scoring in consequential assessment settings.