An Attention-Based Approach for Mongolian News Named Entity Recognition

Abstract

In the field of Natural Language Processing (NLP) of Mongolian, Named Entity Recognition (NER) has great significance. The traditional model is to use the Conditional Random Field (CRF) and Long-Short Term Model (LSTM) method. According to the characteristics of Mongolian, a named entity recognition method based on attention mechanism is proposed in this paper. According to the characteristic of the word-building of the Mongolian language, the suffix of the partial word is divided into morphemes. Based on morphemes, character vectors are trained by LSTM. After that, the word vector is sent to another LSTM to get its context representation. Then the attention mechanism is used to obtain the representation of the full text range of the character vector. Finally, the label sequence of the article is obtained by using CRF. The experimental results show that the Mongolian Named Entity Recognition of attention mechanism is superior to the traditional Bi-LSTM-CRF joint model.