Corpus Construction for Etiquette Entity and Relation Extraction in Ancient Chinese Texts Assisted by Large Language Models

To address the issue of data scarcity in entity relationship extraction tasks within the domain of ancient texts and meet the needs of researchers in digital humanities, this paper proposes an automated data annotation method based on the large language model ChatGPT and prompt engineering. By treating the large language model as a “data annotator,” we developed an entity relationship extraction dataset tailored to ritual scenarios in ancient texts for the natural language processing field. This dataset includes five entity types and ten relationship types, characterized by precise labeling. To verify the robustness and versatility of the dataset, comprehensive experiments were conducted on three pre-trained language models. The results demonstrate that pre-trained models based on prompt learning achieve the best performance in entity relationship extraction tasks. This study shows that the data annotation method leveraging large language models provides a feasible solution for automated data annotation, and the corpus of etiquette entity relationship extraction lays an important foundation for the construction of etiquette question answering system and knowledge graph of ancient books.