Improving relation extraction by multi-task learning
Weijie Wang, Wenxin Hu · 2020
Relation extraction is a subtask of information extraction. Current relation extraction methods are mainly designed for relation extraction tasks, and they use limited knowledge. In this paper, we propose a relation extraction method based on multi-task learning. It uses multiple tasks to learn features that are hard to learn from the relation extraction task, and we add knowledge distillation to help the multi-task model perform better than its single-task counterparts. The experiments based on the pre-trained language model BERT show that our method performs better than most relation extraction methods on the SemEval2010-task8 dataset.