星空之歌：運用 AI 多模態技術打造沉浸式天文感官教學教材
講者/Speaker： 萬信賢，連育賢, WAN HSIN HSIEN , Lien,yu hsien
本作品以「教師賦能與備課」為核心，針對國小自然科學領域中較為抽象、名詞繁雜且關聯複雜的天文單元（涵蓋恆星、視星等、四季大三角與北極星尋找等 16 項核心課綱知識點），提出一套「多模態 AI 教材協作工作流」。本研究旨在解決傳統天文教學缺乏學科整體圖像、短期記憶易疲乏的痛點，並降低一線教師跨足詞曲創作與影像製作的技術門檻。本設計之核心創新點在於建立「文本（課綱知識） $rightarrow$ 音樂（Suno/GarageBand） $rightarrow$ 影像（Canva） $rightarrow$ 課堂落地」的產出模型。設計原則嚴格強調「科學正確性優先於娛樂性」，由教師擔任最後把關者，運用大語言模型（LLM）進行逐句校對（如區分光年為距離單位非時間、校正視星等正負值與亮度關係），並建立提示詞（Prompt）版本管理以利教學資產傳承。音樂結構採主副歌重複核心概念，搭配 Rap 密集輸入操作要點，並結合實體星座盤操作，支援「導入、講解、互動、複習」的全歷程教學策略。教學實踐成效顯著。量化數據顯示，教材影片發布後獲得全台跨校師生廣泛觀看，其平均觀看時間高達 3 分 08 秒，顯著超越影片原生長度（2 分 44 秒），實證了課堂教學引導的停頓互動，以及學習者極高的完整觀看率。質化回饋則指出，本教材成功轉化枯燥知識為高感官吸引力的媒介，有效激發學生自主學習動機，達成「在快樂中主動吸收」的深層學習成效。本工作流亦具備高遷移性，未來可望推廣至其他自然科學單元，為生成式 AI 輔助學科備課與沉浸式教學提供具體可行之範式。關鍵字： 生成式 AI、多模態工作流、天文教學、音樂記憶法、教材正確性、教師賦能
Centered on the core philosophy of "teacher empowerment and lesson preparation," this study proposes a multimodal AI-collaborative workflow tailored for elementary school natural science educators. Astronomy units (covering 16 core curriculum concepts such as stars, apparent magnitude, seasonal triangles, and Polaris navigation) are traditionally abstract and heavily loaded with terminology, often leading to cognitive fatigue among students. To bridge this gap, this project aims to lower the technical entry barriers to songwriting, music arrangement, and video production for frontline teachers, while transforming dry scientific facts into high-quality, reusable multi-sensory materials.The primary innovation lies in the established pipeline: Text (Curriculum Standard) $rightarrow$ Audio (Suno/GarageBand) $rightarrow$ Video (Canva) $rightarrow$ Classroom Implementation. Crucially, the design principle mandates that "scientific accuracy precedes aesthetics." Teachers act as the ultimate gatekeepers, utilizing Large Language Models (LLMs) to perform line-by-line verification—such as distinguishing light-years as units of distance rather than time, and accurately mapping the inverse relationship of apparent magnitude. Furthermore, a strict prompt version control system was implemented to preserve these digital teaching assets. Structurally, the learning materials combine repetitive verse-chorus patterns with rhythmic rap sections for dense procedural knowledge, reinforced by hands-on planisphere (star chart) activities.The instructional outcomes demonstrate both high engagement and educational efficacy. Quantitative results show that the developed materials have been widely adopted across schools in Taiwan. Remarkably, the average watch time reached 3 minutes and 08 seconds, surpassing the actual video length of 2 minutes and 44 seconds. This statistical anomaly strongly indicates that educators frequently paused the video during sessions to introduce scientific concepts, and that learners maintained high completion rates. Qualitative feedback from the comment sections reveals that the integration of auditory and visual stimuli successfully fostered autonomous learning, enabling students to absorb rigorous scientific knowledge joyfully. This workflow exhibits high scalability, offering a reproducible framework for multimodal curriculum design across other scientific domains.Keywords: Generative AI, Multimodal Workflow, Astronomy Education, Musical Mnemonics, Scientific Accuracy, Teacher Empowerment
- 論文全文：星空之歌：運用 AI 多模態技術打造沉浸式天文感官教學教材
- 影像： 相關影片