具思維鏈推理與動機誘因設計之智慧助教環境建置與成效評估;Development and Evaluation of an Intelligent Tutoring Environment Incorporating Chain-of-Thought Reasoning and Extrinsic Motivation Design

NCU Institutional Repository > 資訊電機學院 > 資訊工程研究所 > 博碩士論文 > Item 987654321/98375

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: https://ir.lib.ncu.edu.tw/handle/987654321/98375

题名:	具思維鏈推理與動機誘因設計之智慧助教環境建置與成效評估;Development and Evaluation of an Intelligent Tutoring Environment Incorporating Chain-of-Thought Reasoning and Extrinsic Motivation Design
作者:	陳冠傑;Chen, Kuan-Chieh
贡献者:	資訊工程學系
关键词:	科學模擬;智慧代理人;思維鏈推理;外在動機;science simulation;intelligent agent;Chain-of-Thought reasoning;extrinsic motivation
日期:	2025-07-28
上传时间:	2025-10-17 12:42:06 (UTC+8)
出版者:	國立中央大學
摘要:	科學學習強調學生主動建構知識與理解科學概念，電腦模擬因具備視覺化與高互動性的特質，已廣泛應用於支持學生在科學活動中進行觀察與推理。隨著教學逐漸轉向線上環境，教師難以即時掌握學生的理解狀況與學習歷程，導致學生在操作模擬時常面臨概念迷失與學習瓶頸。為解決此問題，本研究建置一套結合大型語言模型與模擬平台的智慧助教系統，透過提示詞設計與多模態互動，協助學生在科學學習歷程中獲得即時與個別化的學習支援。本研究建構的智慧助教，搭配CoSci電腦模擬平台，打造具備思維鏈推理能力與結構化提示詞框架的教學對話模組。系統可根據學生的操作、模擬畫面與對話內容進行即時判斷與教學引導，並在特定情境下啟動動作模組與評分模組，以回應學生的學習需求。研究以臺灣北部高中一年級學生為對象，共22人，進行一日內的科學學習活動，並設計具外在動機機制（積分條）與無動機設計兩種版本，比較其對學生表現與互動歷程的影響。研究結果顯示，缺乏外在動機機制的學生反而展現出更顯著的學習成效，智慧助教能有效促進其主動提問、批判性思考與知識整合等深層學習行為。此外，智慧助教的回應品質與推理能力亦受提示詞設計影響，具情境整合與概念導向的提示詞更能提升其教學判斷的精準性與互動表現，進而有助於學生的知識建構與理解深化。分析也發現，學生在互動過程中展現的知識處理行為與智慧助教所推論的理解，部分與其學習成果具有顯著關聯。最後根據系統現有限制提出技術改進建議，作為未來智慧助教應用於科學教育之參考。;Science learning emphasizes students’ active knowledge construction and conceptual understanding. Due to their visual and interactive nature, computer simulations have been widely used to support students′ observation and reasoning in scientific activities. However, as instruction increasingly shifts to online environments, it becomes difficult for teachers to monitor students’ understanding and learning processes in real time, often leading to conceptual confusion and learning bottlenecks during simulation-based learning. To address this issue, this study developed an intelligent tutoring system that integrates a large language model with a simulation platform. By incorporating prompt design and multimodal interaction, the system provides students with immediate and individualized learning support throughout their science learning process. The intelligent tutor, built upon the CoSci simulation platform, features a dialogue module equipped with Chain-of-Thought reasoning and a structured prompt framework. The system responds to students’ simulation operations, screen visuals, and dialogues to deliver real-time instructional guidance. When appropriate, it also activates action and assessment modules to address students’ learning needs. This study was conducted with 22 first-year senior high school students in northern Taiwan who participated in a one-day science learning activity. Two interface versions were compared—one with an external motivational mechanism (a score bar) and one without—to examine their impact on students’ performance and interaction behaviors. The results revealed that students without the external motivational mechanism demonstrated significantly better learning outcomes. The intelligent tutor effectively fostered deep learning behaviors such as student-initiated questioning, critical thinking, and knowledge integration. Furthermore, the quality of the tutor’s responses and reasoning was influenced by prompt design; prompts that integrated contextual information and conceptual guidance enhanced the accuracy of instructional judgments and improved interaction quality, thereby supporting students’ knowledge construction and conceptual understanding. Analysis also indicated that students’ knowledge-processing behaviors and the tutor’s inferred understanding were partially correlated with their learning outcomes. Finally, this study proposes technical improvements to address current system limitations and offers recommendations for future applications of intelligent tutors in science education.
显示于类别:	[資訊工程研究所] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	13	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....