Joint Extraction of Entity Relations in Electric Power Industry Standard Based on GCN

The name, author, drafting unit, standard number and other entities of the power standard text can be extracted by the rules. Normative short sentences in the standard text need knowledge extraction to build a knowledge graph. The existing entity relationship joint extraction methods do not fully consider the complex structural characteristics of entity relationships in sentences. Therefore, a new entity relationship joint extraction method based on graph convolution neural network (GCN) is proposed. On the basis of bidirectional long short memory network to extract sequence features, an end-to-end Chinese entity relationship joint extraction model is constructed using the syntax structure information in the GCN encoding dependency analysis results and the improved entity annotation strategy. The experimental results show that the F1 value of this method can reach 72.3%, which is 2.7% higher than that of LSTM-CRF model. GCN can effectively encode the prior word relationship of text and improve the performance of entity relationship extraction.