News text named entity Recognition based on BI-LSTM-CRF model

Abstract

In order to solve the problem of poor discrimination of entity words in Chinese news texts by traditional statistical methods, relying heavily on artificial embedding features and poor model generalization ability, this paper uses Bi-LSTM-CRF (Bi-directional-Long Short Term Memory) based named entity recognition model. The model does not rely too much on manual labeling and rule making of domain knowledge. The model uses bidirectional long sequence processing mechanism of BI-LSTM to understand the context semantics of news text, and uses sequence labeling method of BIO to use Conditional Random Field (CRF) module to constrain the output of text prediction sequence to ensure model recognition ability. In the demonstration experiment, the data set of People's Daily news in 1998 reached the F1 value of 94.53%, which is better than the traditional HMM(Hidden Markov Model) and Conditional Random Field (CRF) models.