Scholarlink

<p>With the deepening of educational evaluation reform, the comprehensive literacy assessment of students presents multidimensional, comprehensive, and highly heterogeneous characteristics. The traditional fixed sampling mode is difficult to adapt to dynamic changes and group differences, and building a precise and efficient sampling system has become the key. This study constructs a multi-stage sampling hierarchical system based on Markov decision processes and introduces an improved strategy gradient optimization algorithm to achieve adaptive iteration. The model integrates historical information weighting, multi-objective utility function, and adaptive learning rate to jointly optimize sampling error, evaluation reliability and validity, and sample resources. On the basic education quality monitoring dataset, the model has the lowest average sampling relative error of 0.82%, the highest confirmatory factor analysis comparative fit index (CFI, as construct validity) of 0.947, and the highest sample size optimization rate of 37.6%. Its comprehensive performance is better than the four mainstream models. In extreme scenarios, the attenuation of evaluation reliability is controlled within 3.2%, and the response efficiency of dynamic indicator adjustment scenarios exceeds 95%. This technological path is suitable for regional monitoring, school-based evaluation, and dynamic tracking assessment.</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

A multi-stage adaptive sampling model based on policy optimization and its application in holistic student competency assessment

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月21日

查看全文
  • Yuze Nie,
  • Ling Yu,
  • Yuting Hu,
  • Xiao Ma

摘要

With the deepening of educational evaluation reform, the comprehensive literacy assessment of students presents multidimensional, comprehensive, and highly heterogeneous characteristics. The traditional fixed sampling mode is difficult to adapt to dynamic changes and group differences, and building a precise and efficient sampling system has become the key. This study constructs a multi-stage sampling hierarchical system based on Markov decision processes and introduces an improved strategy gradient optimization algorithm to achieve adaptive iteration. The model integrates historical information weighting, multi-objective utility function, and adaptive learning rate to jointly optimize sampling error, evaluation reliability and validity, and sample resources. On the basic education quality monitoring dataset, the model has the lowest average sampling relative error of 0.82%, the highest confirmatory factor analysis comparative fit index (CFI, as construct validity) of 0.947, and the highest sample size optimization rate of 37.6%. Its comprehensive performance is better than the four mainstream models. In extreme scenarios, the attenuation of evaluation reliability is controlled within 3.2%, and the response efficiency of dynamic indicator adjustment scenarios exceeds 95%. This technological path is suitable for regional monitoring, school-based evaluation, and dynamic tracking assessment.