Knowledge graph can effectively integrate multi-source data in the power system,improve the level of grid knowledge management.In light of the scarcity of power datasets,diverse entity types and strong professionalism,a method for power data entity recognition based on enhanced optimization pre-trained language model is proposed.This method utilizes data augmentation techniques based on entity word bags to expand the original dataset,employs enhanced optimization pre-trained language model for dynamic semantic encoding,and utilizes bidirectional long short term memory networks and conditional random fields to extract features and optimize labels.Experimental results demonstrate that this entity recognition method outperforms traditional deep learning-based entity recognition methods by 2.17%in F1 score,its effectiveness is confirmed in constructing knowledge graphs for power data.
关键词
知识图谱/实体识别/数据增强/预训练语言模型/双向长短期记忆网络/条件随机场
Key words
knowledge graph/entity recognition/data augmentation/pre-trained language model/bidirectional long short term memory network/conditional random field
维普
万方数据