Sequence Generation with Label Augmentation for Relation Extraction

Sequence generation demonstrates promising performance in recent information extraction efforts, by incorporating large-scale pre-trained Seq2Seq models. This paper investigates the merits of employing sequence relation extraction, finding that with names or synonyms as targets, their textual semantics and correlation (in terms word pattern) among them affect model performance. We then propose Relation Extraction Label Augmentation (RELA), a automatic label augmentation for RE. By saying augmentation, we mean prod semantically each name target. Besides, present an in-depth analysis model's behavior when dealing Experimental results show RELA achieves competitive compared previous methods on four RE datasets.