Person Entity Recognition in Literary Text using BERT

In recent years, the development of transformer-based models such as Bidirectional Encoder Representations from Transformers (BERT) has notably affected the field of Natural Language Processing (NLP). The BERT and its different variants (RoBERTa, DistilBERT, and ALBERT) are proficient in understanding the meaning of text, which is achieved by encoding the input sequence and generating contextual embeddings. The BERT-based models have already impacted the Name Entity Recognition (NER) by providing a powerful way to understand context and improving accuracy. This work used the BERT-based models in comparative analysis to find the best model that recognizes the characters mentioned in literary texts, such as stories and novels. We have developed an NER dataset by annotating the sentences collected from the texts of different literary writings. This dataset has been used to fine-tune the selected BERT-based models. Experimental result of these models shows that RoBERTa-base outperforms the other BERT-based models in person entity recognition in literary text with a training F1-score of 0.895 and a testing F1-score of 0.882. As far as we know, no prior work has addressed using the BERT-based models to recognize the characters mentioned in literary texts. Identifying or recognizing the characters in literary texts helps researchers in the field of literature to investigate and analyze narrative writings. This method also helps in social network construction from literary text, which can be used in summarization and analyzing literary writings to extract meaningful information.