Construction methodology of entity corpus for special diseases electronic medical records

Addressing the issue of resource scarcity for named entity recognition tasks in the medical field, a unified annotation methodology for special diseases entity corpora was formulated under the guidance of medical experts, and two special diseases entity corpora were constructed, namely Pediatric Bronchopneumonia Entity Corpus and Diabetes Entity Corpus. To verify the effectiveness of the proposed special disease entity corpus annotation method, the Pediatric Bronchopneumonia Entity Corpus was first compared with the publicly available dataset using BERT-BiLSTM-CRF and ERNIE-BiLSTM-CRF models. Then, the methodology was reapplied to diabetes electronic medical records to evaluate the robustness of the model. The results showed that both special diseases entity corpora got higher F1 scores than the public datasets, which suggests that special diseases entity corpus annotation methodology proposed in this paper has good robustness.

Key Words: electronic medical record; named entity recognition; corpus construction; Pediatric Bronchopneumonia Entity Corpus; Diabetes Entity Corpus