Creating Knowledge Graph of Electric Power Equipment Faults Based on BERT–BiLSTM–CRF Model

Creating a large-scale knowledge graph of electric power equipment faults will facilitate the development of automatic fault diagnosis and intelligent question answering (QA) in the electric power industry. However, most existing methods have lower accuracy in Chinese entity recognition, thus it is hard to build such a high-quality knowledge graph by extracting knowledge from Chinese technical literature. To solve the problem, a novel model called BERT–BiLSTM–CRF is proposed. It blends Bi-directional Encoder Representation from Transformers (BERT), Bi-directional Long Short-Term Memory (BiLSTM), and Conditional Random Field (CRF). The model firstly identifies and extracts electric power equipment entities from pre-processed Chinese technical literature. Then, the semantic relations between the entities are extracted based on the relation classification method based on dependency parsing. Finally, the extracted knowledge is stored in the Neo4j database in the form of the triplet and visualized in the form of a graph. Through the above steps, a Chinese knowledge graph of electric power equipment faults can be built. The novelty of the model just lies in its subtle blend: the BERT module can not only learn phrase-level information representation, but also learn rich semantic information features; the CRF module realizes the constraint on the label prediction value and reduces the irregular recognition rate, so the accuracy rate of entity recognition is improved. Taking the Chinese technological literature, which is about fault diagnosis of electric power equipment as the experimental object, the experimental results show that the model identifies and extracts Chinese entities more accurately than traditional methods. Thus, a comprehensive and accurate Chinese knowledge graph of electric power equipment faults could be constructed more easily.