Construction of Power Knowledge Graph based on Entity Relation Extraction

Abstract

In order to integrate the fragmented text data in the power domain and solve the problems of disordered and weak correlation of transmission protocols, an improved BERT model was proposed by combining deep learning and knowledge graph for entity relationship extraction in the power domain. This method uses the BERT model based on a full word mask to generate sentence vectors, word vectors with contextual semantics, and then takes the average value of word vectors to get entity vectors. The sentence vectors and entity vectors are combined by the attention machine. Finally, the combined new vectors are put into a fully layer for sequential labeling and finding the optimal tag to implement the entity extracted object. The experimental results show that the precision, recall value, and F1 score of this method are 90.12%, 85.25%, and 87.56 % respectively when entity extraction is performed on the corpus data set of transmission procedures.