A Named Entity Recognition Method for Fields Including Electrical Power Based on Data Enhancement

Deep learning models have been widely applied and continuously developed in named entity recognition tasks. However, the training of most named entity recognition models relies on a large number of annotated corpora. The number of accurately annotated datasets is relatively small, which limits the further development of Chinese named entity recognition models in professional fields such as electrical power. In order to solve the problem of data scarcity, data augmentation technology has become a commonly used strategy. In this paper two data augmentation methods for Chinese named entity recognition tasks based on partial semantic transformation and random noise injection, aiming to make these methods more suitable for the semantic environment of Chinese, are proposed. The proposed method effectively increases the amount of training data and improves the performance of the model on small-scale datasets in specialized fields such as power where resources are limited. This study provides technical support for the potential value of electrical text data.