Supervised NRE models suffer from the lack of large-scale high-quality training data, since manually labeling data is time-consuming and human-intensive. To alleviate this issue, distant supervision (DS) assumption has been used to automatically label data by aligning existing KGs with plain text.