Document-Level Relation Extraction Based on Fine-Grained Information Guidance

Chujun Pu,Xuejie Zhang,Jin Wang,Xiaobing Zhou
DOI: https://doi.org/10.1007/978-981-97-5669-8_31
2024-01-01
Abstract:Document-level Relation Extraction aims to extract relations between entity pairs from complex multi-sentence documents, posing a challenging task due to its complexity. In this paper, we propose a semi-supervised relation extraction model guided by fine-grained information to address issues such as underutilization of entity information and document noise. This approach effectively guides the model to capture context and entity information relevant to entity pairs in relation extraction. Firstly, we introduce an entity multi-granularity information module to effectively utilize entity information and facilitate interaction between entities. Secondly, we propose a module that guides attention using fine-grained evidence to enhance evidence retrieval performance and generate more comprehensive context embeddings for entity pairs. Finally, by generating pseudo-labels and evidence silver labels, we use them as two supervision signals to overcome the noise introduced by distant supervision data. We conduct experiments on two general-domain datasets of DocRE, and the results demonstrate that our model achieves state-of-the-art performance in both relation extraction and evidence sentence retrieval.
What problem does this paper attempt to address?