Augmentation of Learning Content With Knowledge Components: Automatic Unit Labeling for Various Forms of Japanese Math Materials

As the use of learning contents on educational platforms increases, it is desirable to augmentation these contents with unit information representing skills and knowledge in accordance with the curriculum. However, in many cases there is a heavy burden placed on domain matter experts to manually label the unit information to such contents. Against this background, the demand for automatic labeling of unit information to learning contents is increasing. In previous research, classification using n-gram and random forest yielded high performance for automatic unit labeling. These findings were only found for homogeneous learning contents because the method analyzed common words in the contents text. In this study, we conducted an experiment to find the best-performing methods that can be used to label unit information in various forms of textual math learning contents. The experimental result showed that a perceptron method using bigram as a vectorization method performed well for all combinations of prediction datasets. Our proposed method outperforms when labeling contents even in situations where merely a small number of different types of learning contents are available. Implementation of this system will enable the analysis of student behaviors from a content-based perspective, with assisting teachers in efficiently organizing uploaded materials by unit, and with helping students identify relevant content for targeted review.