Emphasis on Easy Samples for Distantly Supervised Relation Extraction

Anonymous

Abstract: There are many wrongly-labeled samples and low-quality samples in automatically generated Distantly Supervised Relation Extraction datasets. Overfitting these samples leads to decline of generalization. To address this issue, the learning of high-quality samples should be prioritized. In this paper, we propose the Emphasis on Easy Samples (EES) mechanism to emphasize high-quality samples using weight distribution regularization at sentence level and priority weighting at bag level. Experiments on a widely used benchmark show that our approach achieves significant improvements.

Paper Type: short