An Enhanced Method for Entity Trigger Named Entity Recognition Based on POS Tag Embedding

In the task of Named Entity Recognition, plenty of human annotations are required in deed. However, a large number of annotations in articles are time-consuming and labor-intensive. In order to solve these problems above, an enhanced method for entity trigger named entity recognition based on POS tag embedding is proposed in this paper. Firstly, by employing lexical annotation tool, it can not only obtain the POS tag of the word, but also connect the word embedding with the POS tag embedding. Secondly, train the attention representation of sentences and triggers, and learn the semantic relationship between entity triggers and sentences based on the attention model. Lastly, the model is instructed with a new sentence attention representation as the input of the CRF (Conditional Random Fields) network. The simulation experiments explicate that the proposed can expand the semantic information of words, so as to improve the recognition ability of entities in a relatively small amount of labeled training data.