Multilevel Electricity Text Named Entity Classification Based on Enhanced XLNet Algorithm

Named Entity Recognition (NER) holds significant importance in natural language processing as it entails identifying and categorizing named entities within text. Within the domain of electricity, precise identification and classification of named entities assume a pivotal role across diverse applications. To efficiently achieve named entity recognition in electrical text, this paper introduces an enhanced XLNet model that integrates XLNet, Bidirectional Long Short-Term Memory (Bi-LSTM), and Conditional Random Field (CRF). XLNet is a Transformer-based pre-trained model that effectively captures long-range dependencies and contextual information by using an autoregressive method and permutation language modeling. Bi-LSTM is capable of capturing the forward and backward dependencies in the text, aiding in extracting contextual features of entities. CRF models the dependencies between labels, leading to improved performance in sequence labeling tasks. Additionally, the paper incorporates a hierarchical fine-tuning process to achieve multilevel entity classification, addressing the specific requirements of the electricity domain.