RoGraphER: Enhanced Extraction of Chinese Medical Entity Relationships Using RoFormer Pre-Trained Model and Weighted Graph Convolution

• DOI: 10.3390/electronics13152892
• Publication Date: 2024-07-23T18:26:50Z
• Authors: Qinghui Zhang; Yaya Sun; Pengtao Lv; Lei Lu; Mengya Zhang; Jinhui Wang; Chenxia Wan; Jingping Wang
• Keywords: Convolution (computer science)

Unstructured Chinese medical texts are rich sources of entity and relational information. The extraction relationships from is pivotal for the construction knowledge graphs aiding healthcare professionals in making swift informed decisions. However, these presents a formidable challenge, notably due to issue overlapping relationships. This study introduces novel model that leverages RoFormer’s rotational position encoding (RoPE) technique an efficient implementation relative encoding. approach not only optimizes positional information utilization but also captures syntactic dependency by constructing weighted adjacency matrix. During feature fusion phase, employs attention mechanism deeper integration features, effectively addressing challenge Experimental outcomes demonstrate our achieves F1 score 83.42 on datasets featuring relations, significantly outperforming other baseline models.