Abstract: To address the issues of heterogeneous content,high complexity,and inability to recognize contextual semantics in diabetic patient-patient online health communities,the BERT-embedded BiLSTM-CRF model is used to recognize six types of entities from the text of the diabetic community "Sweet Home":disease,clinical manifestation,medicine,test,body parts,and emotion.The results demonstrate that the model achieved an accuracy rate of 90.8%,a recall rate of 76%,and a F₁ value of 82.8%.The BERT-BiLSTM-CRF model has been shown to be useful for recognition and to serve as a benchmark for more personalized health information and services.

Key words: diabetes mellitus, online health community, named entity recognition, BERT model