1.NARI Group Corporation (State Grid Electric Power Research Institute), Nanjing 211106, China;2.China Realtime Database Co., Ltd., Nanjing 211106, China

Considering the characteristics of small scale, nested entities and abbreviated entities for electric corpus, the named entity recognition (NER) method based on enhanced text features is proposed. Firstly, by the way of the low-grain word segment and the preset dictionary, the semantic information in Chinese words is properly utilized, and the transmission errors caused by word segment are decreased. Secondly, the structure features of a single Chinese word are learned by the word-level bidirectional gated recurrent unit (Word BiGRU). Together with the features of the part of speech for words and word length, the enhanced word vector is built by concatenating these feature vectors with word vectors. Finally, the NER model is designed with BiGRU, attention mechanism and conditional random field (CRF). The proposed method is verified by using electric corpus and F1 score reaches 87.02%, which proves the effectiveness of NER for electric power industry.