Research on power entity recognition technology base on BiLSTM-CRF

With the increasing demand for data automation and intelligent management in the field of electric power, knowledge atlas plays a particularly important role in intelligent query, electric power knowledge management and decision support. Entity recognition is the basis of building power knowledge map. In the field of power grid, because of its many types, large numbers and strong professionalism of entities, it is difficult to extract them by rules or manually, and it is also impossible to directly reuse entity recognition methods in other fields. For the field of power, this paper first analyzes the subject and characteristics of entity recognition, defines the task objectives of entity recognition, and then studies the entity recognition methods and technologies, and combines the advantages of BiLSTM and CRF, and proposes a method of power entity recognition based on BiLSTM CRF. Firstly, bi-directional features of text are extracted based on BiLSTM, and then the best sequence mark is obtained based on CRF method as the segmentation result. Based on the corpus of power science and technology papers, a comparative experiment is carried out. The accuracy rate of identifying power entities is about 88%, which is 3%~10% higher than the current common methods, which proves that this method can effectively identify power entities.