Network analysis and sentiment classification of Minnan nursery rhymes

Abstract

Minnan nursery rhymes (MNRs), an integral part of Minnan intangible cultural heritage (ICH), are shared by southern Fujian, Taiwan, and overseas Chinese communities. Preserving and analyzing MNRs, especially their emotional evolution over time, is crucial but challenging due to shifting cultural contexts. Traditional sentiment analysis methods often overlook intrinsic relationships among nursery rhymes. To address this, we construct an MNR network using textual feature vectors and cosine similarity thresholds, enabling the exploration of structural and emotional patterns. Inspired by graph neural networks, we propose a Joint-GraphSAGE model for sentiment classification, which effectively captures complex relationships and semantic nuances. Comparative experiments with classical machine learning and deep learning models demonstrate that the Joint-GraphSAGE model significantly outperforms baselines in sentiment classification tasks for both traditional and modern MNRs. This approach not only enhances sentiment analysis accuracy for MNRs, but also offers new perspectives for studying ICH cultural connections.