Named Entity Recognition in the Domain of Geographical Subject

Aiming at Named Entity Recognition (NER) in the domain of geographical subject, two different methods are proposed to recognize two categories of entity: geographical core terms and geographical location. In this paper, conditional random field (CRF) model combining commonly used features and geographical domain features is employed. We additionally perform an extensive number of experiments to verify the effectiveness of geographical domain features and achieve 78.51% and 83.10% in F-1 for two categories of entity respectively. We also propose multi-channel convolutional neural network (MCCNN) which utilizes word embedding features and word base features. A simple MCCNN with little hyperparameter tuning and one layer of convolution achieves excellent results on NER task and achieves 83.15% and 89.92% in F-1 for two categories of entity respectively.