UniER: A Unified and Efficient Entity-Relation Extraction Method with Single-Table Modeling

Sha Liu, Dongsheng Wang, Yue Feng, Miaomiao Zhou, Xuewen Zhu

Abstract: Joint entity and relation extraction are crucial tasks in natural language processing and knowledge graph construction. However, existing methods face several challenges. Firstly, they lack effective interaction modeling between entities and entity relations. Secondly, they often overlook the correlation between entities and relations. Lastly, the utilization of complex models results in increased memory usage. To tackle these challenges, we propose a unified approach that utilizes a homogeneous representation to model the interaction between subject and object entities, as well as entity relations, through a novel method of utilizing a single two-dimensional table for joint extraction tasks. This approach effectively handles multiple token entities, leading to faster and more efficient extraction. Our proposed approach achieves performance comparable to state-of-the-art models on two widely-used datasets and outperforms existing methods on the SEO and HTO problems, which are particularly challenging in relation extraction. It exhibits significant advantages in terms of both speed and space consumption. Our code and models are available at https://anonymous.4open.science/r/UniER.

External IDs:dblp:conf/nlpcc/LiuWFZZ23