自然科協作概念圖評改回饋系統
講者/Speaker： 張文良，霍建豪，鄭鴻哲, Chang Wen-Liang 、Huo Chien-hao、Cheng Hung-che
協作概念圖的評估為教育工作者帶來重大挑戰，特別是在識別個人貢獻水準和提供及時、個人化回饋方面。傳統的協作學習評估方法往往無法準確區分個別學生的貢獻，導致不公平的評估結果和過重的教師工作負擔。隨著生成式人工智慧技術的快速發展，特別是大型語言模型在教育評估中的應用，為解決這些挑戰提供了新的技術路徑。
本研究旨在：(1)透過AI提示系統構建並驗證能夠準確分析協作概念圖並識別個別學生貢獻的框架；
(2)建立客觀的計算機制來量化協作概念圖活動中的個人貢獻值；
(3)設計並實施基於個人貢獻分析的個人化回饋系統，提供差異化學習建議。
我們採用設計導向研究方法，對象為台灣三個班級的90名八年級學生，構建了專門針對自然科學概念圖評估的「角色-任務-步驟-規則」(RTSR)提示優化框架。研究使用NotebookLM作為主要AI評估平台，因其在處理PDF格式協作概念圖的批次處理能力方面表現優異。研究開發了五個版本的提示優化框架，從基礎評估(V1.0)發展到個人化回饋(V5.0)，系統性地提升了系統的評估準確性和回饋品質。
驗證結果顯示系統具有高度可靠性，個人貢獻分析在識別顏色編碼學生貢獻方面達到94%的準確率。貢獻值計算模型有效量化了四種不同貢獻類型的協作參與量和質。個人化回饋機制提供了符合學生認知發展水準的個別化學習建議，顯著提升了協作概念圖活動的教育價值。本研究成功建立了一個全面的AI輔助協作概念圖評估系統，不僅解決了傳統評估方法的限制，也為個人化教育提供了新的技術路徑。研究發現對促進智慧教育發展和提升協作學習效果具有重要的理論和實務價值。
關鍵字： AI輔助評量、協作概念圖、提示詞優化、個人貢獻值、差異化回饋
Assessment of collaborative concept mapping poses significant challenges for educators, particularly in identifying individual contribution levels and providing timely, personalized feedback. Traditional collaborative learning assessment methods often fail to accurately distinguish individual student contributions, leading to unfair assessment outcomes and excessive teacher workload. With the rapid development of generative artificial intelligence technology, particularly the application of large language models in educational assessment, new technical pathways have emerged to address these challenges.
This study aims to: (1) construct and validate a framework capable of accurately analyzing collaborative concept maps and identifying individual student contributions through AI prompt systems; (2) establish objective computational mechanisms to quantify personal contribution values in collaborative concept mapping activities; (3) design and implement personalized feedback systems based on individual contribution analysis, providing differentiated learning recommendations.
We adopted a design-based research approach involving 90 eighth-grade students from three classes in Taiwan, constructing a "Role-Task-Step-Rule" (RTSR) prompt optimization framework specifically for natural science concept map assessment. The study used NotebookLM as the primary AI assessment platform due to its excellent batch processing capabilities for PDF-format collaborative concept maps. Five versions of the prompt optimization framework were developed, evolving from basic assessment (V1.0) to personalized feedback (V5.0), systematically enhancing the system's assessment accuracy and feedback quality.
Validation results demonstrate high system reliability, with individual contribution analysis achieving 94% accuracy in identifying color-coded student contributions. The contribution value calculation model effectively quantified collaborative participation quantity and quality across four different contribution types. The personalized feedback mechanism provided individualized learning recommendations aligned with students' cognitive developmental levels, significantly enhancing the educational value of collaborative concept mapping activities.This study successfully established a comprehensive AI-assisted collaborative concept map assessment system that not only addresses limitations of traditional assessment methods but also provides new technical pathways for personalized education. The research findings hold significant theoretical and practical value for promoting intelligent education development and enhancing collaborative learning effectiveness.
Keywords: AI-assisted assessment, collaborative concept mapping, prompt optimization, individual contribution value, differentiated feedback
- 論文全文：自然科協作概念圖評改回饋系統
- 影像： 相關影片