Abstract: 【 Purpose/significance】 The online health community contains a large amount of complex consultation texts, which require the comprehensive use of multiple types of text semantic representation methods to obtain the texts' semantic information, in order to accurately identify the named entities of the texts.【 Method/process】 Firstly, Word2vec and GloVe static word embedding models are used to train the texts to obtain local and global static semantic representations respectively, which are further fused as a static seman⁃ tic representation. Secondly, apply the ERNIE-Health dynamic word embedding model in the health field to generate dynamic seman⁃ tic representations. Finally, the static and dynamic semantic representations are fused as input of BiLSTM-CRF to fully learn the se⁃ mantic dependencies of the texts' context and the dependency relationships between entity labels, to identify entities of the texts.【 Re⁃ sult/conclusion】 Compared to the single static semantic representation and dynamic semantic representation, the named entity recogni⁃ tion of the fused semantic representation improves accuracy, recall, and F1 values by 2.17%、4.07% and 3.12%, as well as 0.60%、 3.18% and 1.89%, respectively.【 Innovation/limitation】 Comprehensively considering the static and dynamic semantic representation in the consultation texts can improve the named entity recognition of the texts. However, there are still some entities with the same meaning but different names, which affects the quality of the entities in certain degree.