Person Relation Extraction from Ancient Chinese Celebrity Biographies Using BERT-BiGRU-Attention Model

This paper explores the complexity and diversity of expressing personal relationships in ancient Chinese regional celebrity biographies, which primarily stem from historical and cultural differences. These differences lead to issues such as ambiguity, polysemy, and a lack of standardization, thus increasing the difficulty of automated extraction. The implicit nature of personal relationships and their multiple interpretations further exacerbate this challenge. To address these issues, this paper proposes a model combining BERT, BiGRU, and the Attention mechanism for the task of person relation extraction from ancient Chinese regional celebrity biographies. The experiment is based on biographies of celebrities from 12 prefecture-level cities in Zhejiang Province, covering nearly two thousand years of history from the Eastern Zhou to the Qing dynasty, spanning 12 different dynasties. A corpus containing 2,000 annotated sentences was constructed, and five main types of personal relationships were identified: mentor-student, friendship, kinship, political, and others. The experimental results show F1 scores of 0.91, 0.95, 0.956, 0.92, and 0.913 for the five relationship types, demonstrating the model's effectiveness and robustness in handling the complexities of personal relationships in ancient biographies. This lays a foundation for the construction of knowledge graphs for historical figures.