近幾年教學環境應用新興的技術，如人工智能和大數據分析等技術，以更有效地處理和分析教育數據，這些技術的應用為教學團隊提供了更準確和個性化的評估方式，使他們能夠將研究焦點關注在學生的動機和學習策略，以利深入瞭解學生常用的學習方法以及對學習成效的影響，這種趨勢反映出教育領域對於提升學習成效的重視，並透過運用先進技術的能力來瞭解學生的需求和學習狀況。 過去許多研究透過問卷調查和學習序列分析探討其對學習成效的影響，如何考慮到序列的完整性以及建立使用者對工具分析結果的信任是當前教育研究中重要的議題，因此，本實驗旨在比較SBERT+K-means序列分群和傳統序列分群方法間的效果，同時，我們使用可解釋AI的SHAP方法來解釋模型對學習序列使用情形的預測結果，並討論問卷調查和學習序列的使用頻率對學習成效的影響，透過這些分析，我們可以深入了解特徵對模型決策的重要性，以解釋學生的動機和策略以及他們使用的學習行為對學習成效的貢獻。 實驗結果顯示，SBERT結合K-means分群方法的效果能更清晰地呈現出不同群體間的潛在關係，接著透過SHAP方法解釋MSLQ量表預測學習序列使用的影響，觀察到擅長推理和深入思考的學生在閱讀電子書時可能不常使用Highlight和Notetaking等方式進行學習，此外，儘管學生學習時地焦慮程度較高，但他們因為在填寫問卷時表示自己具有較好的動機和策略行為，因此仍持續進行學習以理解知識。實驗進一步探討，問卷調查和學習序列的結合對於學習成效的影響，發現學習序列的使用在學習成效的預測任務中佔有相當大的影響力，這顯示預測學習成效時學習序列的使用是關鍵的因素。;In recent years, emerging technologies such as artificial intelligence and big data analytics have been applied in educational environments to handle and analyze educational data more effectively. These applications provide instructional teams with more accurate and personalized assessment methods, enabling them to focus their research on student motivation and learning strategies. This helps us better understand the learning methods that students commonly use and how these methods affect their learning outcomes. This trend shows that researchers are focusing on improving learning performance and using advanced technologies to understand what students need and how they are learning. Many studies have explored the effect of learning performance through questionnaire surveys and learning sequence analysis. In current educational research, it is important to make sure the sequence is complete and to ensure that users trust the results of the analysis. Therefore, this experiment aims to compare the effectiveness between SBERT+K-means sequence clustering and traditional sequence clustering methods. Additionally, we employ the interpretable AI method called SHAP to explain the predictive results of the model regarding the usage of learning sequences. Furthermore, we discuss the impact of questionnaire surveys and the frequency of learning sequence usage on learning performance. Through these analyses, we can understand the importance of features in model decisions, explaining the contributions of student motivation, strategies, and learning behaviors to learning outcomes. The results indicate that using SBERT and K-means clustering method can more clearly demonstrate the underlying relationships among different groups. Furthermore, SHAP method was employed to explain the influence of MSLQ scales on predicting the usage of learning sequences. It was observed that students who are proficient in analytical thinking and deep thinking may not frequently employ learning methods such as highlighting and notetaking while reading e-books. Additionally, despite experiencing higher levels of anxiety during learning, these students continue to engage in learning activities to comprehend the knowledge due to their reported motivation and strategic behavior in the questionnaire. The experiment further explores the combined impact of questionnaire surveys and learning sequences on learning outcome. It was found that the usage of learning sequences holds considerable influence in predicting learning performance, highlighting the critical role of learning sequence usage in predicting learning performance.
