Chinese Relation Extraction Using Lattice GRU

Most of the existing Chinese entity relation extraction models have adopted methods which are based on character or word input. However, the word-based input method ignores potential word information in the sentence, and the method based on word input will be affected by the spread of word segmentation errors. In order to solve these problem, we introduce a lattice model that can merge words and characters input and use BiGRU instead of BiLSTM. After comparative experiments, our model achieves better results.