[20a-P09-1~6] 23.1 Joint Session N "Informatics"

[20a-P09-5] Recognition of polymer property names in tables by machine learning

〇Hiroyuki Oka¹, Masashi Ishii¹ (1.NIMS)

Keywords:polymer data, property name recognition, machine learning

Recognizing polymer property names describing in tables was examined using machine learning of documents using BERT. Since table captions are effective to recognize the property names, they were also used. The training data were prepared using the sentences formed by connecting the property name and the table caption with a semicolon, and the training was performed using sentence classification task. As a first examination, the expressions for glass transition temperature were trained, giving the F1 scores of 0.93 or higher.