머신러닝 활용 초등영어 교과서 텍스트 군집화 및 그림책 텍스트 분류 연구

This study explores how machine-learning technique can enhance text selection in primary English education by clustering textbook dialogues according to their communicative functions and by classifying picture-book texts within the same framework. 484 dialogue texts from five publishers’ Grade 3~6 textbooks were vectorized and clustered with an unsupervised algorithm. 13 clusters aligned perfectly with a single communcative functions, while 32 clusters showed high concordance when second-most frequent function was also considered. The validated cluster labels served as training data in a logistic-regression classifier that assigned seven English picture books to curriculum-specified communicative functions. Four picture books were classified with high probability for a single function, confirming that models trained on textbook dialogues can credibly map out-of-textbook reading materials onto the curriculum. Educationally, the approach furnishes an objective tool for teachers to identify picture-book texts that reinforce the communicative goals of a given unit. More broadly, it demonstrates that quantitative text analytics can reveal latent connections between mandated textbooks and external resources, offering a scalable pathway toward more coherent and diversified input in primary English classrooms.