A Named Entity Recognition Method for Substation Location Selection
Aigang Cao, Zhenchang Wang, Yongjie Ye, Tiefeng Zhang · 2023
Aiming at the Named Entity Recognition (NER) problem in the field of substation location selection, this paper takes the self-built substation location dataset as the research object, and proposes a NER method combining data enhancement and comparative learning. Firstly, the back-translation method and dropout in RoBERTa are adopted for data enhancement, then the contrastive learning method is used to pre-train the model to improve the robustness of the model. After pre-training, the deep learning model RoBERTa-BiLSTM-CRF is utilized for NER. In the RoBERTa-BiLSTM-CRF model, RoBERTa is first used to obtain the semantics, attribution and location information of words to enhance the semantic representation ability of word vectors. Then the word vector sequence is input into the BiLSTM layer to obtain context information and extract long-distance features. Finally, CRF is used to restrict the legality of sequence tags. The calculation example results indicate that the Pre, Recall and F1 of the proposed method are high, reaching 85.23%, 83.27% and 84.24%, which verifies the effectiveness of the proposed approach.