基于多智能体与混合检索的学科核心素养表现性评价设计研究  

Research on the Design of Performance Assessments for Disciplinary Core Competencies Based on Multi-Agent Systems and Hybrid Retrieval

作　　者：王永固 刘泉[2] 李晓娟 余泽宇[2] Wang Yonggu;Liu Quan;Li Xiaojuan;Yu Zeyu(College of Education,Zhejiang University of Technology,Hangzhou Zhejiang 310014;Mental Health Education Center,Zhejiang University of Finance and Economics,Hangzhou Zhejiang 310018)

机构地区：[1]浙江工业大学教务处 [2]浙江工业大学教育学院,浙江杭州310014 [3]浙江财经大学心理健康教育中心,浙江杭州310018

出　　处：《远程教育杂志》2025年第2期31-44,共14页Journal of Distance Education

基　　金：国家自然科学基金面上项目“多模态特征融合的自闭症教育机器人情感社交智能感知模型及应用研究”(项目编号:62177043);国家社会科学基金项目“新时代大学生社会主义核心价值观知行转化机理及促进策略研究”(项目编号:20BKS099)的研究成果。

摘　　要：在全球教育发展进程中,基于核心素养的智能化教育评价已然成为主流趋势。其中,表现性评价作为衡量学生高阶思维能力的有效手段,于当下教育评价体系中占据关键地位。然而,当前表现性评价在实践层面暴露出诸多问题,主要体现为课程标准导向性欠缺、真实情境构建不充分以及评价规则推理效能低下。针对上述问题,研究基于概念性测评框架(CAF)理论,融合多智能体协同推理与混合RAG策略,构建了一套集评价目标提取、任务设计和量规制定于一体的智能化表现性评价设计系统。为验证该系统的有效性,研究选取中等职业学校信息技术课程中的三个典型教学单元展开实验评估。结果显示:其一,系统生成的评价目标文本在语义准确度和内容完整度方面均表现优异;其二,在任务设计维度,系统在真实性、多样性以及目标一致性上成效显著;其三,在量规设计方面,系统在核心素养导向性和高阶思维递进性维度显著优于教师组设计成果。研究创新性地实现了多智能体与混合RAG策略的深度融合,构建了证据驱动的智能化表现性评价设计范式,为提升核心素养评价设计的科学性与效率提供了新理论路径和创新技术方案,对推动学科核心素养智能化评价的深入发展,具有重要的理论价值与实践指导意义。Intelligent assessment centered on core competencies has become a prevailing trend in global education reform.Among various approaches,performance assessment has emerged as a critical method for evaluating students’higher-order thinking skills and plays a pivotal role in contemporary evaluation systems.However,current practices face several challenges,including insufficient alignment with curriculum standards,limited incorporation of authentic contexts,and low efficiency in the reasoning of assessment rules.In order to address these issues,this study proposes an intelligent performance assessment design system grounded in the conceptual assessment framework(CAF).The system integrates multi-agent collaborative reasoning with a hybrid retrieval-augmented generation(RAG)strategy,enabling the automated design of assessment objectives,task scenarios,and evaluation rubrics within a unified framework.An experimental study was conducted to validate this system and selected three representative teaching units from the Information Technology curriculum in secondary vocational schools.The results indicate that:(1)the system-generated assessment objectives achieved high levels of semantic accuracy and content completeness;(2)in task design,the system demonstrated strong performance in terms of authenticity,diversity,and goal alignment;(3)and in rubric development,the system significantly outperformed teacher-generated rubrics in both core competency alignment and higher-order thinking progression.This study offers a novel,evidence-driven paradigm for performance assessment design through the deep integration of multi-agent systems and hybrid RAG strategies.It provides both theoretical insights and practical solutions for enhancing the scientific rigor and efficiency of assessments targeting disciplinary core competencies,contributing to the advancement of intelligent evaluation practices in competency-based education.

关 键 词：以证据为中心 核心素养 表现性评价 多智能体 检索增强生成 

分 类 号：G420[文化科学—课程与教学论]