Domain Named Entity Recognition Method Based on Skip-gram Model

Traditional domain named entity recognition (NER) methods mainly depended on manual features and were implemented by machine learning methods. These features have no capability to express semantic meaning and these methods are very sensitive for artificial features. To resolve these problems, a method based on Skip-gram model is proposed in this paper. In this method, using word embedding with semantic meaning as features, named entity recognition problem is straightly modeled as Skip-gram model, so it achieves end-to-end solution. Domain characteristics are integrated into this model for further improvement in result. The experiment is carried on Sogou and domain corpus. It shows that the proposed method can improve Recall, Precision and F measure of domain named entity recognition.