This study adopted a novel text mining (TM) technique in curriculum studies to analyze the multi-layered South Korean curriculum document (CD) system from 2012 to 2017. A total of 716 CDs from the national, regional, and school levels corresponding to 23.4 million Korean characters were examined through keyword frequency analysis, topic modeling with Latent Dirichlet Allocation (LDA), construction of text network with clusters, and document vectorization. The results reveal the decentralized nature of each CD layer; i.e., the regional CD (RCD) mediates the national CD (NCD) to the school CD (SCD), and the SCD plays a key role in the implementation of diverse student-centered learning experiences. However, when the similarity (reversely distance) between the average CD vectors of each layer was examined, the SCD showed a constant trend of recentralization toward NCD. This study suggests that the adoption of TM techniques in curriculum studies would enhance the methodological progress of this field in accordance with conventional CD understanding, revealing novel aspects with reproducible and consistent results from a much larger dataset.