現行多數教育聊天機器人僅具應答功能，缺乏對學生學習歷程的整合與呈現，教師若欲掌握學生的理解狀態，需額外閱讀對話紀錄並進行分析，致使對話數據難以有效作為教學決策依據。為協助教師即時掌握學生於課程中的知識狀態，本研究提出一套結合對話知識追蹤與教材結構分析的AI輔助系統，能自動化從學生對話中推估知識掌握程度，並以結構化教材圖譜進行視覺化呈現。

本研究運用LightRAG分析教師所提供之教材內容，自動生成具知識點脈絡的知識圖譜，並結合大型語言模型（LLM）建構具階層與主題結構的心智圖，形塑作為知識追蹤基礎的教材架構。創新地將知識追蹤機制整合於教育聊天機器人的對話過程中，系統能根據教材節點自動標註學生對各知識點的掌握情況，並提供即時追蹤報告協助聊天機器人進行個人化引導，以雷達圖等多元視覺化方式，協助教師快速比較學生在各主題下的學習成效。

本研究以大學國文課程之實際教學場域為實驗對象，蒐集學生與AI助教對話資料，並將系統自動評分結果與教師人工評分進行比較。實驗結果顯示，系統於特定情境下能有效反映學生知識掌握狀態，且其評分結果與教師判斷具顯著相關性。進一步透過虛擬學生實驗驗證，即時知識追蹤機制能有效提升教育聊天機器人的引導效果，特別對被動學習風格學生的改善最為顯著。透過調整追蹤報告內容，系統成功實現「鞏固已知」與「擴展新知」兩種教學策略的應用，證實了個人化教學引導的可行性。綜合而言，本研究不僅提出一套整合教材分析、對話理解與學習歷程可視化的流程，更創新地將知識追蹤技術應用於教育聊天機器人的即時對話中，為對話式AI在智慧教育領域的應用提供具體參考。

Most educational chatbots focus solely on responding to students and lack tools for integrating and visualizing learning trajectories, leaving teachers to manually analyze conversation records to understand students' comprehension states. This makes dialogue data difficult to effectively utilize as a basis for instructional decision-making. To help teachers promptly grasp students' knowledge states during courses, this study proposes an AI-assisted system that combines dialogue-based knowledge tracing with instructional content analysis, capable of automatically inferring students' knowledge mastery from dialogue interactions and presenting results through structured material graphs.

This research employs LightRAG to analyze teacher-provided instructional materials, automatically generating knowledge graphs with conceptual contexts, and combines large language models (LLMs) to construct hierarchical and thematic mind maps, forming the foundation for knowledge tracing. Innovatively integrating knowledge tracing mechanisms into the dialogue process of educational chatbots, the system can automatically annotate students' mastery of knowledge points based on material nodes and provide real-time tracking reports to assist chatbots in personalized guidance, while using diverse visualization methods such as radar charts to help teachers quickly compare students' learning effectiveness across different topics.

This study uses a university Chinese literature course as the experimental setting, collecting dialogue data between students and AI teaching assistants, and comparing automated system scoring with manual teacher evaluations. Experimental results show that the system effectively reflects students' knowledge mastery states in specific contexts, with scoring results significantly correlating with teacher judgments. Further validation through virtual student experiments demonstrates that real-time knowledge tracing mechanisms can effectively enhance the guidance effectiveness of educational chatbots, with the most significant improvements observed for passive learning style students. By adjusting tracking report content, the system successfully implements two teaching strategies: "consolidating known knowledge" and "expanding new knowledge," confirming the feasibility of personalized instructional guidance. In summary, this research not only proposes an integrated workflow for material analysis, dialogue understanding, and learning trajectory visualization, but also innovatively applies knowledge tracing technology to real-time dialogue in educational chatbots, providing concrete references for conversational AI applications in smart education.

[1] Mohammad Amin Kuhail, Nazik Alturki, Salwa Alramlawi, andKholood Alhejori. Interacting with educational chatbots: A system-atic review. Education and Information Technologies, 28(1):973–1018, 2023.
[2] Siyu Wu, Yang Cao, Jiajun Cui, Runze Li, Hong Qian, Bo Jiang,and Wei Zhang. A comprehensive exploration of personalized learn-ing in smart education: From student modeling to personalized rec-ommendations, 2024.
[3] Alexander Scarlatos, Ryan S. Baker, and Andrew Lan. Exploringknowledge tracing in tutor-student dialogues using llms, 2024.
[4] National Governors Association et al. Common core state stan-dards. Washington, DC, 2010.
[5] Guozheng Li, Peng Wang, and Wenjun Ke. Revisiting large lan-guage models as zero-shot relation extractors, 2023.
[6] Zirui Guo, Lianghao Xia, Yanhua Yu, Tu Ao, and Chao Huang.Lightrag: Simple and fast retrieval-augmented generation, 2025.
[7] 林子平. 基於大型語言模型應用指示詞打造無程式碼對話系統平台 - 以聊故事機器人為例. 碩士論文, 國立中央大學, 2023.
[8] Joseph Weizenbaum. Eliza—a computer program for the study ofnatural language communication between man and machine. Com-munications of the ACM, 9(1):36–45, 1966.
[9] Gwo-Jen Hwang and Ching-Yi Chang. A review of opportunitiesand challenges of chatbots in education. Interactive Learning En-vironments, 31(7):4099–4112, 2023.
[10] Shen Wang, Tianlong Xu, Hang Li, Chaoli Zhang, Joleen Liang,Jiliang Tang, Philip S Yu, and Qingsong Wen. Large languagemodels for education: A survey and outlook. arXiv preprintarXiv:2403.18105, 2024.
[11] Luke K Fryer, Mary Ainley, Andrew Thompson, Aaron Gibson, andZelinda Sherlock. Stimulating and sustaining interest in a languagecourse: An experimental comparison of chatbot and human taskpartners. Computers in human behavior, 75:461–468, 2017.
[12] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and KristinaToutanova. Bert: Pre-training of deep bidirectional transformersfor language understanding. In Proceedings of the 2019 conferenceof the North American chapter of the association for computationallinguistics: human language technologies, volume 1 (long and shortpapers), pages 4171–4186, 2019.
[13] Alec Radford, Jeffrey Wu, Rewon Child, David Luan, DarioAmodei, Ilya Sutskever, et al. Language models are unsupervisedmultitask learners. OpenAI blog, 1(8):9, 2019.
[14] Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah,Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, PranavShyam, Girish Sastry, Amanda Askell, et al. Language models arefew-shot learners. Advances in neural information processing sys-tems, 33:1877–1901, 2020.
[15] Lixiang Yan, Lele Sha, Linxuan Zhao, Yuheng Li, RobertoMartinez-Maldonado, Guanliang Chen, Xinyu Li, Yueqiao Jin, andDragan Gašević. Practical and ethical challenges of large languagemodels in education: A systematic scoping review. British Journalof Educational Technology, 55(1):90–112, 2024.
[16] Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, FeiXia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large language models. Advances inneural information processing systems, 35:24824–24837, 2022.
[17] Luyu Gao, Aman Madaan, Shuyan Zhou, Uri Alon, Pengfei Liu,Yiming Yang, Jamie Callan, and Graham Neubig. Pal: Program-aided language models. In International Conference on MachineLearning, pages 10764–10799. PMLR, 2023.
[18] Wenhu Chen, Xueguang Ma, Xinyi Wang, and William W Co-hen. Program of thoughts prompting: Disentangling computa-tion from reasoning for numerical reasoning tasks. arXiv preprintarXiv:2211.12588, 2022.
[19] Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li,Erkang Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu,et al. Autogen: Enabling next-gen llm applications via multi-agentconversation. arXiv preprint arXiv:2308.08155, 2023.
[20] Ethan Prihar, Morgan Lee, Mia Hopman, Adam Tauman Kalai,Sofia Vempala, Allison Wang, Gabriel Wickline, Aly Murray, andNeil Heffernan. Comparing different approaches to generatingmathematics explanations using large language models. In Inter-national Conference on Artificial Intelligence in Education, pages290–295. Springer, 2023.
[21] Changrong Xiao, Wenxing Ma, Sean Xin Xu, Kunpeng Zhang, Yu-fang Wang, and Qi Fu. From automation to augmentation: Largelanguage models elevating essay scoring landscape. arXiv e-prints,pages arXiv–2401, 2024.
[22] Luoma Ke, Song Tong, Peng Cheng, and Kaiping Peng. Exploringthe frontiers of llms in psychological applications: A comprehensivereview. arXiv preprint arXiv:2401.01519, 2024.
[23] Gautam Yadav, Ying-Jui Tseng, and Xiaolin Ni. Contextualizingproblems to student interests at scale in intelligent tutoring sys-tem using large language models. arXiv preprint arXiv:2306.00190,2023.
[24] Jing Liang, Ryan Hare, Tianyu Chang, Fangli Xu, Ying Tang, Fei-Yue Wang, Shimeng Peng, and Mingyu Lei. Student modeling andanalysis in adaptive instructional systems. IEEE Access, 10:59359–59372, 2022.
[25] Jaime R Carbonell. Ai in cai: An artificial-intelligence approach tocomputer-assisted instruction. IEEE transactions on man-machinesystems, 11(4):190–202, 2007.
[26] Neil T Heffernan and Cristina Lindquist Heffernan. The assist-ments ecosystem: Building a platform that brings scientists andteachers together for minimally invasive research on human learn-ing and teaching. International Journal of Artificial Intelligence inEducation, 24(4):470–497, 2014.
[27] Li Cai, Kilchan Choi, Mark Hansen, and Lauren Harrell. Itemresponse theory. Annual Review of Statistics and Its Application,3(1):297–321, 2016.
[28] Richard M Felder. Learning and teaching styles in engineering ed-ucation. 2002.
[29] Ran Liu and John C Stamper. Multimodal data collection andanalysis of collaborative learning through an intelligent tutoringsystem. In MMLA-CrossLAK@ LAK, pages 47–52, 2017.
[30] Aya Nabil, Mohammed Seyam, and Ahmed Abou-Elfetouh. Predic-tion of students’academic performance based on courses’gradesusing deep neural networks. IEEE Access, 9:140731–140746, 2021.
[31] Minju Park, Sojung Kim, Seunghyun Lee, Soonwoo Kwon, andKyuseok Kim. Empowering personalized learning through aconversation-based tutoring system with student modeling. In Ex-tended Abstracts of the CHI Conference on Human Factors in Com-puting Systems, CHI ’24, page 1–10. ACM, May 2024.
[32] Albert T Corbett and John R Anderson. Knowledge tracing: Mod-eling the acquisition of procedural knowledge. User modeling anduser-adapted interaction, 4:253–278, 1994.
[33] Ghodai Abdelrahman, Qing Wang, and Bernardo Nunes. Knowledge tracing: A survey. ACM Computing Surveys, 55(11):1–37,2023.
[34] Michael V Yudelson, Kenneth R Koedinger, and Geoffrey J Gordon.Individualized bayesian knowledge tracing models. In Artificial in-telligence in education: 16th international conference, AIED 2013,Memphis, TN, USA, July 9-13, 2013. proceedings 16, pages 171–180. Springer, 2013.
[35] Jill-Jênn Vie and Hisashi Kashima. Knowledge tracing machines:Factorization machines for knowledge tracing. In Proceedings of theAAAI conference on artificial intelligence, volume 33, pages 750–757, 2019.
[36] Chris Piech, Jonathan Bassen, Jonathan Huang, Surya Ganguli,Mehran Sahami, Leonidas J Guibas, and Jascha Sohl-Dickstein.Deep knowledge tracing. Advances in neural information processingsystems, 28, 2015.
[37] Alex Sherstinsky. Fundamentals of recurrent neural network (rnn)and long short-term memory (lstm) network. Physica D: NonlinearPhenomena, 404:132306, 2020.
[38] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit,Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin.Attention is all you need, 2023.
[39] Hiromi Nakagawa, Yusuke Iwasawa, and Yutaka Matsuo. Graph-based knowledge tracing: modeling student proficiency using graphneural network. In IEEE/WIC/aCM international conference onweb intelligence, pages 156–163, 2019.
[40] Zhangqi Duan, Nigel Fernandez, Arun Balajiee Lekshmi Narayanan,Mohammad Hassany, Rafaella Sampaio de Alencar, PeterBrusilovsky, Bita Akram, and Andrew Lan. Automated knowledgecomponent generation and knowledge tracing for coding problems,2025.
[41] Tuan Bui, Oanh Tran, Phuong Nguyen, Bao Ho, Long Nguyen,Thang Bui, and Tho Quan. Cross-data knowledge graph construc-tion for llm-enabled educational question-answering system: A casestudy at hcmut. In Proceedings of the 1st ACM Workshop on AI-Powered Q&A Systems for Multimedia, ICMR ’24, page 36–43.ACM, June 2024.
[42] Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni,Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis,Wen tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela.Retrieval-augmented generation for knowledge-intensive nlp tasks,2021.
[43] Tianyu Gao, Xingcheng Yao, and Danqi Chen. Simcse: Sim-ple contrastive learning of sentence embeddings. arXiv preprintarXiv:2104.08821, 2021.
[44] Nils Reimers and Iryna Gurevych. Sentence-bert: Sen-tence embeddings using siamese bert-networks. arXiv preprintarXiv:1908.10084, 2019.
[45] Chang Liu, Loc Hoang, Andrew Stolman, and Bo Wu. Hita: A rag-based educational platform that centers educators in the instruc-tional loop. In International Conference on Artificial Intelligencein Education, pages 405–412. Springer, 2024.
[46] Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, AlexChao, Apurva Mody, Steven Truitt, Dasha Metropolitansky,Robert Osazuwa Ness, and Jonathan Larson. From local to global:A graph rag approach to query-focused summarization, 2025.
[47] Qinggang Zhang, Shengyuan Chen, Yuanchen Bei, Zheng Yuan,Huachi Zhou, Zijin Hong, Junnan Dong, Hao Chen, Yi Chang, andXiao Huang. A survey of graph retrieval-augmented generation forcustomized large language models, 2025.
[48] Rui Yang, Boming Yang, Xinjie Zhao, Fan Gao, Aosong Feng, SixunOuyang, Moritz Blum, Tianwei She, Yuang Jiang, Freddy Lecue,et al. Graphusion: A rag framework for scientific knowledge graphconstruction with a global perspective. In Companion Proceedingsof the ACM on Web Conference 2025, pages 2579–2588, 2025.
[49] Yuxin Yang, Haoyang Wu, Tao Wang, Jia Yang, Hao Ma, and Guo-jie Luo. Pseudo-knowledge graph: Meta-path guided retrieval andin-graph text for rag-equipped llm, 2025.
[50] Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, BrianIchter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. Chain-of-thought prompting elicits reasoning in large language models, 2023.
[51] 黃冠傑. 輕鬆打造教師的代理人：基於 eduact 的聊天機器人創建與系統模組改善。. 碩士論文, 國立中央大學, 2024.
[52] Yang Liu, Dan Iter, Yichong Xu, Shuohang Wang, Ruochen Xu,and Chenguang Zhu. G-eval: Nlg evaluation using gpt-4 with betterhuman alignment. arXiv preprint arXiv:2303.16634, 2023.
[53] Alexander Scarlatos, Naiming Liu, Jaewook Lee, Richard Baraniuk,and Andrew Lan. Training llm-based tutors to improve studentlearning outcomes in dialogues, 2025.
[54] Yi-Ting Yeh, Maxine Eskenazi, and Shikib Mehri. A comprehensiveassessment of dialog evaluation metrics, 2021.
[55] Menna Fateen and Tsunenori Mine. Developing a tutoring dialogdataset to optimize llms for educational use, 2024.
[56] Troy Camarata and Tony A Slieman. Improving student feed-back quality: A simple model using peer review and feedbackrubrics. Journal of medical education and curricular development,7:2382120520936604, 2020.