Perkembangan Automated Essay Scoring Untuk Bahasa Arab: Systematic Literature Review Dari Machine Learning Hingga Large Language Models

Student writing assessment is an important component of the educational process. However, manual essay assessment is time-consuming and labor-intensive and may be influenced by evaluator subjectivity. Automated Essay Scoring (AES) has been developed as a technology-based solution to support faster and more consistent writing assessment. Nevertheless, the implementation of AES for Arabic still faces various challenges compared with English. These challenges are related to the characteristics of Arabic, including its root-and-pattern morphology, complex word formation processes, and variations in language use and dialects.This article presents a Systematic Literature Review (SLR) examining the development of AES methodologies for Arabic. The review traces the evolution of approaches from manual feature extraction using traditional Machine Learning, through Deep Learning based on Transformer architectures, to recent developments utilizing generative Large Language Models (LLMs). The findings indicate that Transformer-based models demonstrate strong capabilities in understanding relationships between different parts of a text within a broader context. Meanwhile, LLMs such as Gemini, DeepSeek, and Jais offer new opportunities for more comprehensive writing assessment, including vocabulary, writing style, and mechanics.The review also identifies several important developments, including the use of synthetic data through error injection techniques to address the limited availability of annotated Arabic corpora. In addition, prompt relevance modeling can be used to ensure that assessed essays are genuinely aligned with the assigned topic or question. Overall, the development of Arabic AES demonstrates a shift from systems that merely generate numerical scores toward systems capable of functioning as artificial intelligence-based tutors.Such systems have the potential to provide more informative and transparent feedback while adapting to the characteristics of learners' writing.