Preliminary Study of Named Entity Recognition Techniques in Japanese Light Novels

This research presents an initial attempt at computational experiments for Named Entity Recognition (NER) in Japanese light novels, aiming to understand and manage their unique writing styles and vocabularies, with potential future applications for more complex analyses and automated summarization technologies. Specifically, we annotated nine Japanese light novels to construct a foundational dataset and subsequently conducted an experimental study on NER utilizing deep learning models along with various data augmentation techniques. While the NER accuracy achieved in this study is currently limited, we have obtained valuable insights into the unique expressions and stylistic elements present in light novels. These findings are expected to contribute significantly to the advancement of future information extraction techniques.