Ancient Chinese Poetry Collation Based on BERT.

The rapid advancements in intelligent knowledge management technologies, exemplified by generative large language models, have yet to be fully explored and applied in the field of collation of ancient Chinese poetry. This study investigates the application of BERT-based pre-trained models, namely bert-base-chinese and SikuBERT, in the specialized task of ancient Chinese poetry collation. Focusing on the poetry of Li Bai, we employed a meticulously curated dataset to fine-tune these models, with the objective of enhancing their ability to identify and rectify errors in classical verse. Through a systematic approach to model adaptation, our research aimed to bridge the gap between generic language understanding and the nuanced complexities of ancient poetry.Results indicate that both models, after fine-tuning, exhibit substantial improvement in accurately addressing textual issues in the poetry. Specifically, SikuBERT, with its background in classical Chinese literature, achieved an impressive accuracy rate exceeding 40% post-fine-tuning, reflecting a marked increase from its base performance, thereby validating the significance of domain-specific training data. Meanwhile, bert-base-chinese also displayed notable enhancements, underscoring the models' adaptability to specialized tasks. The investigation further emphasizes the potential for artificial intelligence to contribute to the precision and efficiency of ancient literature studies. We highlight future directions including refining fine-tuning methodologies, expanding the models' capability to generalize across diverse poetic styles and periods, and integrating multi-modal data to deepen the understanding of historical context and authorial intent. This work underscores the transformative role of AI in the digital preservation and scholarly analysis of ancient poetry, paving the way for innovative approaches in the field of classical literature collation. [ABSTRACT FROM AUTHOR]