Abstract: As an item on UNESCO's Representative List of Intangible Cultural Heritage of Humanity, Yimakan embodies the historical memory and cultural genes of the Hezhen ethnic group. The annotation and excavation of its texts hold significant academic value for the digital preservation of endangered ethnic minority cultural heritage. To address the issues of low-resource annotated corpora and the ambiguity of culturally specific entity boundaries within Yimakan texts, this research introduces an enhanced RoBERTa-BiLSTM-CRF model for named entity recognition. Comparative experimental evaluations indicate that the proposed model exhibits significantly superior performance over conventional baseline approaches, with its strength particularly evident in fine-grained entity identification. This model not only lays an extensible groundwork for the construction of the Yimakan knowledge graph but also provides valuable methodological inspiration for advancing intelligent text-processing paradigms for intangible cultural heritage.