結合生成式人工智慧與自然語言處理技術之「合作科學探究想法收斂自動回饋鷹架」系統開發與初步評估

Development and Preliminary Evaluation of an Automated Feedback Scaffold for Idea Convergence in Collaborative Scientific Inquiry Integrating Generative AI and Natural Language Processing Technologies

本研究開發了一個基於自然語言處理技術的「想法收斂自動回饋鷹架」系統，旨在促進學生於合作科學探究中的知識翻新與深入學習。透過後設言談（meta-discourse）的應用，該系統於學生合作探究過程中提供結構化引導與反思機制，幫助學生有效組織和檢視其想法，進而加深對學習內容的理解與反思。本研究針對54位小學五年級學生進行為期4週的合作科學探究活動，並結合機器評估（ROUGE-SU9）及一位具有經驗的小學教師進行人工評估，以驗證系統效能。結果顯示，該系統可有效支持學生於合作探究中的知識翻新過程，並提升其探究能力與學習成效。機器評估結果顯示，系統生成的摘要在覆蓋性、相關性及一致性方面均達到較高標準，可準確反映學生的想法並提供有價值的回饋，進一步促進學生進行有效的後設言談與自我反思。教師評估亦支持該系統的實用性，尤其是在支援學生的探究活動及知識翻新方面展現了穩定性與適用性。總結而言，本研究證實了生成式人工智慧及自然語言處理技術在支持合作科學探究及促進知識翻新方面的潛力。未來研究可進一步探討這些技術在不同學科與教育情境中的應用，以提升教學效果，為教育科技創新提供新方向。

This study developed an “Automatic Feedback Scaffold” system based on natural language processing technology, aimed at facilitating students’ knowledge refinement and deep learning during collaborative scientific inquiry. By incorporating meta-discourse, the system provides structured guidance and reflection mechanisms throughout the collaborative inquiry process. It helps students effectively organize and examine their ideas, thereby enhancing their understanding and reflection on the learning content. The study involved 54 fifth-grade elementary school students over a four-week collaborative scientific inquiry activity. Machine evaluation (using ROUGE-SU9) and manual evaluation by an experienced elementary school teacher were conducted to verify the system’s effectiveness. The results indicated that the system effectively supports students’ knowledge refinement processes during collaborative inquiry, while also improving their inquiry abilities and learning outcomes. Machine evaluation results demonstrated that the system-generated summaries achieved high standards in terms of coverage, relevance, and consistency, ensuring the timely delivery of valuable feedback that further promotes effective meta-discourse and self-reflection. Teacher evaluation results also supported the system’s practicality, particularly in its stability and applicability in facilitating students’ inquiry activities and knowledge refinement. In summary, this study confirms the potential of generative artificial intelligence and natural language processing technology to support collaborative scientific inquiry and promote knowledge innovation.