Scholarlink

<p>Automated essay scoring (AES) systems have relied on artificial intelligence (AI) models; from traditional machine learning models to large language models (LLMs). Several systematic literature reviews (SLRs) have been conducted on automated essay scoring (AES) systems in the past. A thorough scan of the literature showed that most SLRs have focused on specific AI techniques/models, but not a comprehensive array of these techniques/models. The purpose of this study is to fill this gap by conducting an SLR on AES systems, specifically from traditional machine learning models to large language models. Essentially, the study tries to capture a comprehensive array of AI models used in AES systems; The Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) methodology is used for the review. An initial set of 473 records from the Scopus, Web of Science, IEEE and ACM Digital Library databases were screened. Forty-three peer-reviewed open access articles were screened at the final stage. While transformer-based and LLM-driven systems often had higher QWK values, improvements were more closely linked to rubric alignment, trait-specific modelling, and construct-aware scoring mechanisms than to model scale alone. During the review period, AES systems clearly got better at being generalisable, easy to understand, and in line with education. There was a move away from holistic scoring toward frameworks that focus on multiple traits and constructs. There was also better cross-prompt robustness and the addition of explainable and collaborative human-AI scoring systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

From Traditional Machine Learning Models to Large Language Models: A Systematic Literature Review of Automated Essay Scoring

  • 期刊论文

  • 发表日期：

    2026年5月9日

查看全文
  • Eli Fianu,
  • Fred Amankwah-Sarfo,
  • Philomina Ofori,
  • John K Amoako,
  • Hickma Sumani

摘要

Automated essay scoring (AES) systems have relied on artificial intelligence (AI) models; from traditional machine learning models to large language models (LLMs). Several systematic literature reviews (SLRs) have been conducted on automated essay scoring (AES) systems in the past. A thorough scan of the literature showed that most SLRs have focused on specific AI techniques/models, but not a comprehensive array of these techniques/models. The purpose of this study is to fill this gap by conducting an SLR on AES systems, specifically from traditional machine learning models to large language models. Essentially, the study tries to capture a comprehensive array of AI models used in AES systems; The Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) methodology is used for the review. An initial set of 473 records from the Scopus, Web of Science, IEEE and ACM Digital Library databases were screened. Forty-three peer-reviewed open access articles were screened at the final stage. While transformer-based and LLM-driven systems often had higher QWK values, improvements were more closely linked to rubric alignment, trait-specific modelling, and construct-aware scoring mechanisms than to model scale alone. During the review period, AES systems clearly got better at being generalisable, easy to understand, and in line with education. There was a move away from holistic scoring toward frameworks that focus on multiple traits and constructs. There was also better cross-prompt robustness and the addition of explainable and collaborative human-AI scoring systems.