課程影片理解程度測量

Abstract

了解學生上課時的理解程度變化有助於提高學習效率。若不想影響課程進行，可使用穿戴式的感測器或相機等設備測量學生的反應、透過機器學習模型來分析資料並估計理解程度。然而和理解程度的變化頻率相比，這類資料集往往包含過少的標註資訊。因此我們提出了能夠處理大量資訊的系統，以有效建立含有充足標註資訊的資料集。此外我們實際利用這個系統收集了一個包含身體姿勢、臉部表情和視線軌跡的資料集，用於訓練機器學習模型及預測理解程度。結果顯示，即便在不使用所有資訊的情況下，測出的理解程度仍有可靠的準確度，充分展現了我們所提出的系統被廣泛採用的可能性。

Measuring students’ mental states, such as their understanding during class, helps im prove learning efficiency. Automatic approaches implement this idea without interrupting the class by sensing students’ reactions through devices such as wearable sensors or cam eras, and applying machine learning models to analyze the data. However, most of the previous works lack adequate annotations of understanding based on students’ reactions compared to the number of concepts conveyed during lessons. In this paper, we propose a scalable framework for efficiently constructing and annotating datasets. Additionally, we have collected a dataset consisting of posture, facial expression, and eye movement features, and benchmarked it for measuring understanding. The results show promising accuracy even in cases where not all features are available, demonstrating the potential for widespread adoption of the proposed framework.