Chinese Named Entity Recognition (CNNER) faces numerous challenges, including the diversity of the Chinese language, the complex representation of mixed Chinese and English characters and symbols in texts, the complexity of the Chinese language itself with long sentences containing multiple entities, and the uneven distribution of named entity classes in actual Chinese scenarios. To address these challenges, we propose a method called CPMFA (Character Pair-based method with Multi-feature representation and Attention mechanism). The CPMFA method predicts predetermined relations between character pairs, facilitating the identification of nested named entities based on these relations. Firstly, the method leverages the pre-trained language model LERT (Linguistically-motivated Bidirectional Encoder Representation from Transformer) and BiLSTM (Bidirectional Long Short-Term Memory) to generate comprehensive and accurate character embeddings. Secondly, it incorporates multi-feature representation to capture complex semantic information and introduces the Pyramid Squeeze Attention (PSA) module to emphasize key features. Finally, the PolyLoss function is integrated into the model training process to tackle the challenge of an imbalanced distribution of entity classes. We employed the DiaKG, Yidu-S4K and Weibo datasets to validate and evaluate the efficacy and adaptability of our method. The F1 obtained by the CPMFA on these three datasets is 83.79%, 72.03%, and 70.39%, in that order. The experimental results illustrate the outstanding performance of the proposed CPMFA method in both general knowledge and Chinese medical domains.