Author: In Sook, Jeong
Annotation: This study proposes a seven-code framework to classify Korean idioms based on cultural themes. A 5-million-token corpus was analyzed using TextRank, and 800 idioms were manually labeled to train a BERT-based multi-label classifier (F1 = 0.87). The 100 most frequent idioms were selected to support teaching practices that link idiom learning with cultural literacy.
Keywords: Korean idioms; cultural codes; machine learning; BERT; TextRank; Korean language education
Pages in journal: 13 - 20
