부산대학교 도서관

The process of research activities in the electric power field produces a large number of articles, patents, and other results, which contain a large amount of information. Named entity recognition is a common method for obtaining textual information, and this study establishes a model that can efficiently recognize named entities of electric power research results. Firstly, the keywords of electric power literature are obtained for preprocessing and organization, and then the annotated dataset of keyword named entities of electric power scientific research is constructed. After that, this paper enhances the ability of the model to obtain contextual information through the Transformer encoder and enriches the expression of textual information by obtaining textual semantic structure information through dependent syntactic analysis. Meanwhile, in order to strengthen the feature extraction ability, the BILSTM-Multi-Head-Attention-CRF model is used to obtain text context information. IDCNN model is also used to make this paper's model acquire the ability to extract text semantic structure information. Finally, ablation experiments are conducted to show that the model proposed in this paper has a better recognition effect compared with the named entity recognition methods in the generic domain.