基于BiLSTM-CRF模型识别细密地址的特定标注方法分析

Analysis of Specific Annotation Methods for Identifying Fine-grained Addresses Based on BiLSTM-CRF model

阐述针对人为地址输入进行识别,通过深度学习将语义相似相近的地址完成实体名识别。以BiLSTM-CRF进行数据集的训练,针对前后文联系确定最佳可能出现地址,将海量地址进行特征提取和分类构造,得出多数F1指标91%以上。模型应用于实际数据中,验证了该方法的有效性。

This paper describes the recognition of human input addresses,using deep learning to recognize entity names of semantically similar addresses.Train the dataset with Bi LSTM-CRF,determine the best possible address based on the previous and subsequent connections,extract features and classify massive addresses,and obtain a majority of F1 indicators above 91%.The model was applied to actual data to verify the effectiveness of the method.