The versatility of large language models (LLMs) leads to an increasing penetration rate and market share of LLMs. Many new LLMs have launched, challenging the selection and evaluation of an appropriate LLM for target applications. In this research, attention is drawn to education as one of the most remarkable areas. Evaluating LLMs using different evaluation metrics is both subjective and objective. Using the advanced search tool in Scopus, 231 relevant studies have been shortlisted for the review of LLM evaluation metrics in education. Some basic analyses, including the number of publications by year, a frequency table of the subject areas, distribution of document types, distribution of source titles, a word cloud summary of the top 50 keywords, and distribution of regions, were provided. It was followed by an in-depth discussion of the top 10 highly cited publications. Then, five challenges in LLM evaluation were analyzed.

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

A Review of Large Language Model Evaluation Metrics in Education

  • Naraphorn Paoprasert,
  • Kwok Tai Chui,
  • Tik Yung Yip,
  • Jiaqi Liu,
  • Kwan Keung Ng

摘要

The versatility of large language models (LLMs) leads to an increasing penetration rate and market share of LLMs. Many new LLMs have launched, challenging the selection and evaluation of an appropriate LLM for target applications. In this research, attention is drawn to education as one of the most remarkable areas. Evaluating LLMs using different evaluation metrics is both subjective and objective. Using the advanced search tool in Scopus, 231 relevant studies have been shortlisted for the review of LLM evaluation metrics in education. Some basic analyses, including the number of publications by year, a frequency table of the subject areas, distribution of document types, distribution of source titles, a word cloud summary of the top 50 keywords, and distribution of regions, were provided. It was followed by an in-depth discussion of the top 10 highly cited publications. Then, five challenges in LLM evaluation were analyzed.