Purpose/Significance Electronic medical records (EMR) of liver cancer contain a large amount of medical knowledge, and most of the knowledge is in the form of unstructured data which is difficult to extract automatically. The research on entity recognition of liver cancer EMR is important in the construction of clinical decision support systems and medical knowledge graphs in the area of liver cancer. Method/Process A named entity recognition (NER) model combined with RoBERTa algorithm and CRF algorithm is built, and the model achieves excellent effect. The real data of self-labeled EMR of liver cancer are used for model training and testing. Result/Conclusion RoBERTa-CRF model is better than other baseline models and has good entity recognition effect.