Amir DN Cohen, Shachar Rosenman, Yoav Goldberg.

Most of the current supervised relation classification algorithms use a single embedding to represent the relation between a pair of entities. We argue that a better approach is to treat the relation classification task as a Span-Prediction problem, similar to Question Answering. We present a span prediction based system for relation classification and evaluate its performance compared to the embedding-based system. We demonstrate that the supervised span prediction objective works significantly better than the standard classification-based objective. We achieve state-of-the-art results on the TACRED, SemEval task 8, and CRE datasets.

Citation

@inproceedings{
cohen2022supervised,
title={Supervised Relation Classification as Two-way Span-Prediction},
author={Amir DN Cohen and Shachar Rosenman and Yoav Goldberg},
booktitle={4th Conference on Automated Knowledge Base Construction},
year={2022}
}