Chinese Relation Extraction Using Extend Softword

In recent years, many scholars have chosen to use word lexicons incorporate information into a model based on character input improve the performance of Chinese relation extraction (RE). For example, Li et al. proposed MG-Lattice in 2019 and achieved state-of-the-art (SOTA) results. However, still has problem loss due its structure, which affects RE. This paper proposes an adaptive method include at embedding layer using lexicon merge all words that match each input-based solve MG-Lattice. The can be combined with other general neural system networks transferability. Experimental studies two benchmark RE datasets show our achieves inference speed up 12.9 times faster than SOTA model, along better performance. experimental results also this BERT pretrained effectively supplement obtained from further improving