Named Entity Recognition of liver cancer data based on Damped Pointer Network and Dynamic Fusion

The electronic medical record (EMR) of liver cancer covers a large amount of key information, including the pathological stage of the tumor, the location of the tumor, and the size of the tumor, which helps doctors quickly understand the patient’s condition and make diagnosis. However, conceptually complex and multi-terminological EMR makes doctors hard to retrieve useful information, which not only leads to low work efficiency, but also may miss the key information of pathological understanding. Named entity recognition (NER) technology can help doctors quickly screen out key entities and improve the efficiency of clinicians. In this study, we proposed two model structures (Damped Pointer Network and Dynamic Fusion) to improve the accuracy and recall rate of entity recognition. The model structures were well fitted the EMR with small size of samples and many technical terms in the form of uncommon words. Experiments showed that the F1-score (harmonic average of accuracy and recall rate) of the model structures proposed in this study was 98.56%, which was 2% higher than other frequently-used recognition technologies like BERT and BERT-CRF.