Enterprise AI Analysis

BERT-CRF with Knowledge Graph for Character Relationship Extraction and Inheritance Genealogy Construction in Cantonese Opera Scripts

This paper constructs an annotated dataset for the field of Cantonese opera based on approximately five million characters of corpus from the Guangdong Cantonese Opera Digital Resource Database and the Guangzhou Library Cantonese Opera Literature Database. It proposes a method for role relationship extraction and lineage construction that integrates pre-trained language models and knowledge graphs. In the entity recognition stage, the BERT-CRF model is used to identify key entities such as characters, roles, and schools. In the relationship extraction stage, a classification framework of entity location labeling and context encoding is introduced to automatically extract relationships such as lineage, fellow students, and collaborations. At the knowledge representation level, TransH is used to vectorize the "Cantonese Opera Lineage Knowledge Graph" to achieve link prediction and lineage completion for complex many-to-many master-apprentice relationships. Experimental results show that the constructed model achieves an overall F1 score of 91.82% on the entity recognition task and 87.31% on the relation extraction task. TransH outperforms TransE and TransR in Hits@1, Hits@10, and MRR metrics. This research has achieved the automated construction of a knowledge graph of Cantonese opera traditions from the text, providing technical support for the visualization and quan-titative research of the lineage network of renowned masters. It has certain reference value for promoting the digital protection and development of intangible cultural heritage in the opera genre.