[15p-P06-1~15] 23 Joint Session N "Informatics" (Poster)

[15p-P06-15] Materials Development with LLM (II): Extraction of NdFeB Magnet Data from US Patents

〇Hiroyuki Oka¹, Masashi Ishii¹ (1.NIMS)

Keywords:

NdFeB magnet,data extraction,large language model

Automatic extraction of fabrication conditions for NdFeB magnets from US patents was examined using a Large Language Model (LLM). Gemma 2 was used as the LLM. The fabrication conditions were extracted using the query which includes the character strings specifying the extraction targets such as ‘sinter-temp’. The extraction accuracy from 11 patents was f1 = 0.72, which is higher than that for the corresponding extraction previously performed using the rule-base methods and machine learnings. This presentation will report the details.