本研究旨在設計並實證驗證一套結合大型語言模型（Large Language Models, LLMs）與模型脈絡協定（Model Context Protocol, MCP）──之智慧教學協作架構
MAS（Multi-Agent Large Language Model Systems）。該架構透過語境同步與策略分工機制，協助教學代理人於多任務情境中協同推理、任務分派與回應生成，以提升教學互動品質與系統穩定性。MALS 架構整合多個具專責功能的代理角色（如：範例產生、內容篩選、錯誤診斷、摘要生成等），並透過共享脈絡管理機制實現低耦合之資訊傳遞與任務協作。
    研究採取多階段混合方法設計，涵蓋系統建模、原型實作與多組實驗實證。實驗分為三組場域：國小學生基礎任務組（實驗一）、大學生 SQL 題組（實驗二）與 Python 題組（實驗三），以檢視 MALS 架構在異質學習背景下之應用潛能。資料分析包括互動歷程變項（如 messageCount、toolResponseCount、contextLinkedToolCount）、自建複合指標（cooperationIndex）、統計檢定（ANOVA、皮爾森相關）與語意主題建模（LDA、LSA），以多層次驗證語境共享、決策精確性與教學回應品質之整合成效。
    研究結果指出，MALS 架構可有效提升學生互動焦點、語言組構深度與工具使用策略，並具備穩定系統負載與因應個別差異之潛力。實驗三進一步整合驗證四項核心假設，說明多代理機制能強化語境協同、提升個人化教學反應、減輕 LLMs 計算負擔，並增進整體教學系統之操作效能與維護彈性。本研究貢獻不僅在於系統性建構多代理教學協作架構，亦提出語言主題建模與互動行為融合之分析途徑，為未來智慧教學設計與教學語料研究提供理論與方法基礎。

This study proposes and empirically validates an intelligent instructional collaboration framework that integrates Large Language Models (LLMs) with a Model Context Protocol (MCP), referred to as the Multi-Agent Large Language Model Systems (MALS). The framework is designed to facilitate collaborative reasoning, task allocation, and response generation among instructional agents across multi-task educational scenarios through mechanisms of context synchronization and strategic role assignment. By incorporating multiple specialized agents—such as those responsible for example generation, content filtering, error diagnosis, and summary production—MALS realizes low-coupling information exchange and task coordination via a shared context management system, thereby enhancing the quality of pedagogical interaction and overall system robustness.

    A multi-phase mixed-method research design was employed, encompassing system modeling, prototype development, and empirical evaluations across three distinct educational settings: (1) foundational tasks with elementary school students (Experiment 1), (2) SQL-based assignments with university students (Experiment 2), and (3) Python programming tasks (Experiment 3). These experimental contexts serve to examine the cross-contextual applicability of the MALS framework in diverse learning environments. Data analysis encompassed interactional metrics (e.g., messageCount, toolResponseCount, contextLinkedToolCount), composite indices (e.g., cooperationIndex), statistical analyses (e.g., ANOVA, Pearson’s correlation), and semantic topic modeling techniques (e.g., LDA, LSA), enabling a multi-level validation of contextual coherence, decision accuracy, and pedagogical response quality.

    Findings demonstrate that the MALS architecture significantly enhances learners’ interactional focus, depth of linguistic construction, and strategic tool usage. The framework also shows promise in balancing system load and accommodating learner diversity. Experiment 3 further substantiates four core hypotheses, illustrating that the multi-agent mechanism strengthens contextual coordination, enhances personalized instructional responses, alleviates computational demands on LLMs, and improves operational efficiency and maintainability of the instructional system.

    The contributions of this research are threefold: (1) the systematic development of a multi-agent instructional collaboration framework, (2) the integration of semantic modeling with interactional behavior analysis, and (3) the provision of both theoretical and methodological foundations for future advancements in intelligent tutoring systems and instructional discourse research.

目錄
第１章 緒論	1
第１節 研究背景	1
第２節 研究動機	3
第３節 研究目的	5
第４節 名詞解釋	7
第５節 研究問題	9
第２章 文獻回顧	10
第１節 教學代理人設計理論與應用發展	10
第２節 大型語言模型於教育應用之發展潛力與實踐挑戰	12
第３節 多代理人系統之設計架構與協作挑戰	14
第４節 模型脈絡協定於多代理語境協同之應用與理論設計	17
第５節 文獻綜合分析與研究假設之提出	19
第３章 系統開發	22
第１節 系統設計原則與總體開發流程	22
第２節 系統核心架構說明與代理模組設計	25
第３節 系統運行邏輯與任務協作流程	30
第４節 MCP 協定整合設計與語境資料同步機制	33
第５節 系統部署環境與整合技術堆疊	36
第６節 範例交互場景與代理協作歷程示例	40
第４章 研究方法	44
第１節 研究流程	44
第２節 研究場域與對象	46
第３節 研究假設與研究問題對應	48
第４節 實驗設計與分析策略	49
第５章 研究結果與討論	51
第１節 第一組(國小組)	51
第２節 第二組（SQL 題組）	55
第３節 實驗三（Python 題組）	63
第６章 結論	67
第１節 研究結論	67
第２節 研究貢獻	70
第３節 研究限制	72
第４節 未來研究方向	74
參考文獻	75
附錄	79
附錄一、 國小、Python組(多代理人組) 各代理人prompt	79
附錄二、 SQL組各代理人prompt	84
附錄三、 python組單代理人prompt	95

 
圖目錄
圖 ３ 1系統總體技術架構圖	24
圖 ３ 2智慧代理模組之內部組成結構	26
圖 ３ 3任務流程控制模組之內部架構	27
圖 ３ 4後端 HTTP 服務模組之組成架構	28
圖 ３ 5語言模型整合模組之內部架構	29
圖 ３ 6 MALS 系統整體任務處理流程圖	30
圖 ３ 7工具調用之標準結構	31
圖 ３ 8本研究教學代理系統之邏輯架構與模組整合圖	36
圖 ３ 9使用者與代理間之任務互動畫面截圖：SQL 查詢生成場景	41
圖 ３ 10 Session 紀錄中之工具代理調用內容與回應資訊封裝（MongoDB 節錄）	42
圖 ４ 1研究流程圖	45
圖 ５ 1全組訊息量與任務量箱型圖	52
圖 ５ 2 國小組操作中斷範例：語句未完成與互動未結案	53
圖 ５ 3 國小學生於任務中產生之非任務性輸入與模糊語意指令	54
圖 ５ 4 高分組語意主題建模分析	60
圖 ５ 5 中分組語意主題建模分析	60
圖 ５ 6 低分組語意主題建模分析	61
圖 ５ 7 Group3 LDA/LSA 分析	64
圖 ５ 8 Group4 LDA/LSA 分析	65

 
表目錄
表 ４ 1研究假設與場域對應表	48
表 ５ 1全組描述性統計	51
表 ５ 2第二組描述性統計-1	55
表 ５ 3第二組描述性統計-2	55
表 ５ 4第二組工具請求量單因子變異數分析	56
表 ５ 5第二組工具請求量描述性統計	56
表 ５ 6第二組工具請求量變異數同質性檢定	56
表 ５ 7第二組作業分數單因子變異數分析	57
表 ５ 8第二組作業分數描述性統計	57
表 ５ 9第二組作業分數變異數同質性檢定	57
表 ５ 10第二組訊息量單因子變異數分析	57
表 ５ 11第二組訊息量描述性統計	57
表 ５ 12第二組訊息量變異數同質性檢定	57
表 ５ 13第二組皮爾森相關分析	58
表 ５ 14實驗三描述性統計	63

Abd-alrazaq, Alaa, Rawan AlSaad, Dari Alhuwail, Arfan Ahmed, Padraig Mark Healy, Syed Latifi, Sarah Aziz, Rafat Damseh, Sadam Alabed Alrazak, and Javaid Sheikh. “Large Language Models in Medical Education: Opportunities, Challenges, and Future Directions.” JMIR Medical Education 9, no. 1 (June 1, 2023): e48291. https://doi.org/10.2196/48291.
Abdelghani, Rania, Yen-Hsiang Wang, Xingdi Yuan, Tong Wang, Pauline Lucas, Hélène Sauzéon, and Pierre-Yves Oudeyer. “GPT-3-Driven Pedagogical Agents to Train Children’s Curious Question-Asking Skills.” International Journal of Artificial Intelligence in Education 34, no. 2 (June 1, 2024): 483–518. https://doi.org/10.1007/s40593-023-00340-7.
Baylor, Amy, and Yanghee Kim. “Simulating Instructional Roles through Pedagogical Agents.” International Journal of Artificial Intelligence in Education 15, no. 1 (January 1, 2005): 95–115. https://doi.org/10.1007/s40593-015-0055-y.
Baylor, Amy L. “Expanding Preservice Teachers’ Metacognitive Awareness of Instructional Planning through Pedagogical Agents.” Educational Technology Research and Development 50, no. 2 (June 1, 2002): 5–22. https://doi.org/10.1007/BF02504991.
Chang, Ching-Yi, I.-Hui Chen, and Kai-Yu Tang. “Roles and Research Trends of ChatGPT-Based Learning: A Bibliometric Analysis and Systematic Review.” Educational Technology & Society 27, no. 4 (October 1, 2024): 471–86. https://doi.org/10.30191/ETS.202410_27(4).TP03.
Chen, Lijia, Pingping Chen, and Zhijian Lin. “Artificial Intelligence in Education: A Review.” IEEE Access 8 (2020): 75264–78. https://doi.org/10.1109/ACCESS.2020.2988510.
Chen, Xieling, Di Zou, Haoran Xie, Gary Cheng, and Caixia Liu. “Two Decades of Artificial Intelligence in Education: Contributors, Collaborations, Research Topics, Challenges, and Future Directions.” Educational Technology & Society 25, no. 1 (2022): 28–47.
Clarebout, Geraldine, and Steffi Heidig (née Domagk). “Pedagogical Agents.” In Encyclopedia of the Sciences of Learning, 2567–71. Springer, Boston, MA, 2012. https://doi.org/10.1007/978-1-4419-1428-6_942.
“Designing and Learning With Pedagogical Agents: An Umbrella Review - Noah L. Schroeder, Robert O. Davis, Eunbyul Yang, 2025.” Accessed June 11, 2025. https://journals.sagepub.com/doi/10.1177/07356331241288476.
Díaz, Brayan, and Miguel Nussbaum. “Artificial Intelligence for Teaching and Learning in Schools: The Need for Pedagogical Intelligence.” Computers & Education 217 (August 1, 2024): 105071. https://doi.org/10.1016/j.compedu.2024.105071.
“Effects of Incorporating a Large Language Model-Based Adaptive Mechanism Into Contextual Games on Students’ Academic Performance, Flow Experience, Cognitive Load and Behavioral Patterns - Minkai Wang, Di Zhang, Jingdong Zhu, Hanjie Gu, 2025.” Accessed June 11, 2025. https://journals.sagepub.com/doi/abs/10.1177/07356331251321719.
Fang, Jian-Wen, Guo ,Xiao-Ge, Meng ,Xiao-Pei, Hwang ,Gwo-Jen, and Yun-Fang and Tu. “Realization of Parasocial Interaction Theory in STEAM Education with an AI Pedagogical Agent: Insights from Learning Performance, Collaboration Tendency, and Problem-Solving Tendency.” Interactive Learning Environments 0, no. 0 (n.d.): 1–24. https://doi.org/10.1080/10494820.2025.2508326.
Ganesan, Vijaya Shankar, and Vinit Mojes. “Revolutionizing Education Using Personalized AI: Multi-Agent Approach.” SSRN Scholarly Paper. Rochester, NY: Social Science Research Network, March 10, 2025. https://doi.org/10.2139/ssrn.5173671.
Guo, Taicheng, Xiuying Chen, Yaqi Wang, Ruidi Chang, Shichao Pei, Nitesh V. Chawla, Olaf Wiest, and Xiangliang Zhang. “Large Language Model Based Multi-Agents: A Survey of Progress and Challenges.” arXiv, April 19, 2024. https://doi.org/10.48550/arXiv.2402.01680.
Kim, Yanghee, and Amy L. Baylor. “Research-Based Design of Pedagogical Agent Roles: A Review, Progress, and Recommendations.” International Journal of Artificial Intelligence in Education 26, no. 1 (March 1, 2016): 160–69. https://doi.org/10.1007/s40593-015-0055-y.
Krishnan, Naveen. “Advancing Multi-Agent Systems Through Model Context Protocol: Architecture, Implementation, and Applications.” arXiv, April 26, 2025. https://doi.org/10.48550/arXiv.2504.21030.
Laak, Kristjan-Julius, and Jaan Aru. “AI and Personalized Learning: Bridging the Gap with Modern Educational Goals.” arXiv, March 21, 2025. https://doi.org/10.48550/arXiv.2404.02798.
Lan, Yu-Ju, and Nian-Shing Chen. “Teachers’ Agency in the Era of LLM and Generative AI: Designing Pedagogical AI Agents.” Educational Technology & Society 27, no. 1 (January 1, 2024): i–xviii. https://doi.org/10.30191/ETS.202401_27(1).PP01.
Lin, Hao-Chiang Koong, Chun-Hsiung Tseng, and Nian-Shing Chen. “Enhancing Programming Education: The Impact of AI-Based Pedagogical Agents on Student Self-Efficacy, Engagement, and Learning Outcomes.” Educational Technology & Society 28, no. 2 (April 1, 2025): 279–94. https://doi.org/10.30191/ETS.202504_28(2).TP02.
Liu, Zhi, Huimin Duan, Shiqi Liu, Rui Mu, Sannyuya Liu, and Zongkai Yang. “Improving Knowledge Gain and Emotional Experience in Online Learning with Knowledge and Emotional Scaffolding-Based Conversational Agent.” Educational Technology & Society 27, no. 2 (April 1, 2024): 197–219. https://doi.org/10.30191/ETS.202404_27(2).RP08.
Maldonado, Diego, Edison Cruz, Jackeline Abad Torres, Patricio J. Cruz, and Silvana del Pilar Gamboa Benitez. “Multi-Agent Systems: A Survey About Its Components, Framework and Workflow.” IEEE Access 12 (2024): 80950–75. https://doi.org/10.1109/ACCESS.2024.3409051.
Mejeh, Mathias, and Martin Rehm. “Taking Adaptive Learning in Educational Settings to the next Level: Leveraging Natural Language Processing for Improved Personalization.” Educational Technology Research and Development 72, no. 3 (June 1, 2024): 1597–1621. https://doi.org/10.1007/s11423-024-10345-1.
“Multimedia Learning with Animated Pedagogical Agents (Chapter 38) - The Cambridge Handbook of Multimedia Learning.” Accessed June 11, 2025. https://www.cambridge.org/core/books/abs/cambridge-handbook-of-multimedia-learning/multimedia-learning-with-animated-pedagogical-agents/B629E8948494D949DB818B4AB31F3B4F.
Sethi, Surbhi Seema, and Kanishk Jain. “AI Technologies for Social Emotional Learning: Recent Research and Future Directions.” Journal of Research in Innovative Teaching & Learning 17, no. 2 (May 27, 2024): 213–25. https://doi.org/10.1108/JRIT-03-2024-0073.
Shi, Yao, Rongkeng Liang, and Yong Xu. “EducationQ: Evaluating LLMs’ Teaching Capabilities Through Multi-Agent Dialogue Framework.” arXiv, April 21, 2025. https://doi.org/10.48550/arXiv.2504.14928.
Shoaib, Muhammad, Nasir Sayed, Jaiteg Singh, Jana Shafi, Shakir Khan, and Farman Ali. “AI Student Success Predictor: Enhancing Personalized Learning in Campus Management Systems.” Computers in Human Behavior 158 (September 1, 2024): 108301. https://doi.org/10.1016/j.chb.2024.108301.
Sikström, Pieta, Chiara Valentini, Anu Sivunen, and Tommi Kärkkäinen. “How Pedagogical Agents Communicate with Students: A Two-Phase Systematic Review.” Computers & Education 188 (October 1, 2022): 104564. https://doi.org/10.1016/j.compedu.2022.104564.
Szeider, Stefan. “MCP-Solver: Integrating Language Models with Constraint Programming Systems.” arXiv, April 6, 2025. https://doi.org/10.48550/arXiv.2501.00539.
Yang, Yingxuan, Huacan Chai, Yuanyi Song, Siyuan Qi, Muning Wen, Ning Li, Junwei Liao, et al. “A Survey of AI Agent Protocols.” arXiv, April 26, 2025. https://doi.org/10.48550/arXiv.2504.16736.
Yusuf, Habeeb, Arthur Money, and Damon Daylamani-Zad. “Pedagogical AI Conversational Agents in Higher Education: A Conceptual Framework and Survey of the State of the Art.” Educational Technology Research and Development 73, no. 2 (April 1, 2025): 815–74. https://doi.org/10.1007/s11423-025-10447-4.
Zielke, Marjorie A., Zakhidov ,Djakhangir, Lo ,Tiffany, Craig ,Scotty D., Rege ,Robert, Pyle ,Hunter, Meer ,Nina Velasco, and Nolan and Kuo. “Exploring Social Learning in Collaborative Augmented Reality With Pedagogical Agents as Learning Companions.” International Journal of Human–Computer Interaction 41, no. 4 (February 16, 2025): 2424–49. https://doi.org/10.1080/10447318.2024.2323280.