An Improved Baseline for Sentence-level Relation Extraction
This work provides an improved baseline for relation extraction, offering strong performance gains for researchers and practitioners working on information extraction tasks.
This paper addresses sentence-level relation extraction, focusing on entity representation and noisy labels. The improved baseline achieves an F1 score of 74.6% on TACRED, outperforming previous state-of-the-art methods, and 91.1% on the refined Re-TACRED dataset.
Sentence-level relation extraction (RE) aims at identifying the relationship between two entities in a sentence. Many efforts have been devoted to this problem, while the best performing methods are still far from perfect. In this paper, we revisit two problems that affect the performance of existing RE models, namely entity representation and noisy or ill-defined labels. Our improved RE baseline, incorporated with entity representations with typed markers, achieves an F1 of 74.6% on TACRED, significantly outperforms previous SOTA methods. Furthermore, the presented new baseline achieves an F1 of 91.1% on the refined Re-TACRED dataset, demonstrating that the pretrained language models (PLMs) achieve high performance on this task. We release our code to the community for future research.