Currently, there is a lack of high-quality annotated datasets in the field of ancient Chinese NER and fewer related studies. Based on the “Comprehensive Mirror for Aid in Government” dataset, we revised the entity annotation specification and reconstructed a CMAG-NER2024 dataset. To solve the problem of missing entity information due to omission in ancient Chinese texts, we propose RIF-NER, an NER model for ancient Chinese, which can effectively fuse contextual information. The model retrieves contexts related to inputs from contextual corpora and realizes more effective fusion of inputs and related contexts through sparse cross-attention and dynamic gating mechanisms.