Chinese Entity Relation Extraction Based on Multi-level Gated Recurrent Mechanism and Self-attention
Zicheng Zhong · 2021
There is the influence of polysemy and segmentation quality on semantic understanding for Chinese entity relation extraction, we propose a Chinese entity relation extraction model based on a multi-level gated recurrent mechanism and self-attention (MGRSA). For alleviating segmentation errors, we design the framework of a multi-level gated recurrent mechanism that unites word-grained information into character-grained information. For reducing polysemy ambiguity, we utilize self-attention on two parts, including word vectors with external semantic knowledge. Compared with other models on two datasets, our model can learn more semantic information in the case of introducing less noise, which shows better advantages and robustness.