Optimized Named Entity Recognition of Electric Power Field Based on Word-Struct BiGRU

Abstract

In view of the characteristics of small corpus size, nested entities and abbreviated entities in electric field, this paper studied the methods of entity recognition in professional field, and proposed a method of entity recognition in electric field based on text vector feature enhancement. Firstly, by using low-intensity and preset word segmentation, the semantic information contained in Chinese words can be used reasonably to avoid the influence of semantic separation between single Chinese character and word on text vectors; Then WS-BiGRU is designed to learn the structural features of words, and a new feature enhancement vector is formed by fusing word vectors, parts of speech and word lengths. Furthermore, the training of entity recognition model is completed by BiGRU-Attention-CRF. For electric corpus, the P value is 86.74% and the R value is 87.3%. The research results can be used in electric dispatching knowledge graph, metadata model, intelligent report query and other business scenarios.