Here is a source code deposit for a Japanese journal paper, "Named Entities Extraction by Citizen Participation and Machine Learning for Making Machine-Readable Old Records of the Edo Period Remaining in Local Communities" Yoshiga, N. et. al., Vol. 63 No. 2, 2022. (Not published yet.)
There are many ancient documents such as business records and testimonials written before the Edo period in Japan. Machine-readable metadata will be one of effective tools for utilizing those records. In cases of materials related to a very small area, in particular, it is necessary to deal with unique expressions restricted in the area. In this study, we set a specific goal to generate machine-readable metadata such as Linked Data from the database of the cataloged articles for the Ogi-han Nikki (business records) from the Edo period. We aim to improve the efficiency in extraction processes of named entities. For this purpose, we employ two methods. The first is effective use of human resources through citizen participation. The second is automated extraction of named entities by machine learning. We show that the proposed method works well even for materials related to a local area.
Source codes shown here are based on Flair.
情報処理学会論文誌「郷土に残存する江戸期古記録の機械可読化を目的とした市民参加および機械学習による固有表現抽出」吉賀夏子, 堀良彰, 只木進一, 永崎研宣, 伊藤昭弘. Vol. 63 No. 2, 2022年2月. に掲載しているGoogle Colaboratoryのソースコードと同じです。