Research on Document-Level Person Relation Extraction in Chinese

EasyChair Preprint 15202

Abstract

This study aims to develop a joint entityrelation extraction framework that can be applied to real-world web data. Addressing the limitations of existing datasets, which are often derived from a single source and primarily focused on sentence-level content, we utilize large language models (such as Gemini and GPT-3.5) to annotate articlelevel content and build a more generalized dataset using Chinese Common Crawl data. To enhance the reliability of annotations and the completeness of entity pair sampling, we employ cross-validation and entity augmentation methods. Additionally, we fine-tune pre-trained models to validate and improve the performance of entity-relation extraction in real-world scenarios.

Keyphrases: 命名實體識別, 文章級關係擷取, 聯合實體關係擷取

@booklet{EasyChair:15202,
  author    = {Min-Chao Hung and Chia-Hui Chang and Chi-Ju Yeh},
  title     = {Research on Document-Level Person Relation Extraction in Chinese},
  howpublished = {EasyChair Preprint 15202},
  year      = {EasyChair, 2024}}