| 摘要: | 在臺灣，持續安靜閱讀（MSSR）已被廣泛應用，但由於師生比例問題，學生在閱讀速度與興趣方面存在顯著差異，使得教師難以有效評估每位學生的閱讀成效。為了解決這一挑戰，本研究結合生成式人工智慧（GenAI）聊天機器人與4F動態回顧循環（Active Reviewing Cycle）理論，開發了一個閱讀學習同伴平臺。該平臺透過OpenAI的Assistants API連接，建立一位能作為學生閱讀學習夥伴的聊天機器人。此系統能檢索學生所閱讀書籍的內容，提供「4F」問題（事實、發現、感覺與未來），從多角度引導學生反思、回顧與延伸，並給予回饋與互動式的回應，進而提升其學習成效。系統亦建置聊天互動歷程分析儀表板，協助教師與學生即時掌握互動行為與聊書情況等資訊。
為了評估生成式AI閱讀同伴的影響及其方法的有效性，本研究在臺灣北部某小學進行為期十週的實驗，招募239名國小學生，並將其分為三組：總結回饋組共82人、即時回饋組共89人以及對照組共68人。總結回饋組於完成整段與學習同伴的對話後統一收到回饋；即時回饋組則於互動過程中即時獲得系統回饋；對照組則接受傳統閱讀教學。實驗期間蒐集各組學生的閱讀理解測驗、詞彙測驗、學習動機與學習策略問卷，以及系統日誌資料，以進行整體效益與行為差異之分析。
研究結果顯示，總結回饋組與即時回饋組在閱讀動機與學習策略兩方面皆有顯著提升，顯示兩種AI介入方式皆能有效支持學習歷程；即時回饋組在閱讀理解與詞彙理解表現上顯著優於總結回饋組與對照組，顯示即時性回饋在促進學習成效方面的優勢。此外，即時提供的評分回饋有助於激發學生反思與修正，強化其自我調整能力。系統所建置之聊天互動歷程分析儀表板，亦能協助教師掌握學習表現差異、進行個別化教學，並支持學生進行學習監控與策略調整。綜合而言，本研究證實結合4F提問設計與回饋機制的生成式AI聊天機器人，具有提升閱讀理解、詞彙能力、學習動機與學習策略的潛力，亦有助於培養學生主動反思與自我調整的學習能力，未來可進一步優化互動設計，拓展其在閱讀教學中的應用可能。
;In Taiwan, the practice of Modeled Sustained Silent Reading (MSSR) has been widely implemented. However, due to the imbalance in the teacher-student ratio, there are significant individual differences in students′ reading speeds and interests, making it difficult for teachers to assess each student′s reading outcomes effectively. To address this challenge, this study integrated a Generative AI chatbot and the 4F Active Reviewing Cycle theory to develop a reading companion platform. The platform connects to OpenAI’s Assistants API to create a chatbot that acts as a reading companion for students. This system can retrieve the content of books students have read and generate “4F” questions (Facts, Feelings, Findings, and Future) to guide students’ reflection, review, and extension from multiple perspectives. It also provides interactive feedback and responses designed to enhance learning outcomes. Additionally, a dashboard is integrated to visualize the interaction history, enabling both teachers and students to monitor conversation behaviors and book-related engagement in real-time.
To evaluate the impact and effectiveness of the Generative AI reading companion, a ten-week experimental study was conducted at an elementary school in northern Taiwan, involving 239 students. Participants were divided into three groups: the Summative Feedback Group (n = 82), the Real-Time Feedback Group (n = 89), and a Control Group (n = 68). The Summative Feedback Group received feedback only after completing the entire conversation with the AI companion, while the Real-Time Feedback Group received feedback during the interaction. The Control Group received traditional reading instruction. During the experiment, data were collected from reading comprehension tests, vocabulary tests, questionnaires on learning motivation and strategies, as well as system log data, to analyze overall learning outcomes and behavioral differences among the groups.
The results of this study indicate that both the summative and real-time feedback groups showed significant improvements in reading motivation and learning strategies, demonstrating that both AI-supported interventions effectively facilitated students’ learning processes. Notably, the real-time feedback group outperformed both the summative feedback and control groups in reading comprehension and vocabulary acquisition, highlighting the advantage of immediate feedback in enhancing learning outcomes. Additionally, the provision of real-time scoring feedback encouraged students to engage in reflection and revision, thereby strengthening their self-regulation abilities. The integrated interaction history dashboard also proved valuable for teachers in monitoring performance differences, delivering personalized instruction, and supporting students in self-monitoring and adjusting their strategies. Overall, this study confirms that a generative AI chatbot embedded with 4F questioning prompts and feedback mechanisms holds strong potential for improving students’ reading comprehension, vocabulary skills, learning motivation, and learning strategies. It also contributes to fostering self-reflection and self-regulated learning, suggesting future directions for optimizing interaction design and expanding its application in reading instruction. |
