[4Rin1] Interactive 2

[4Rin1-56] Graph-based Unsupervised Important Technical Terms Extraction Method for Patent Document

〇Asahi Hentona¹, Hirofumi Nonaka¹, Hidekazu Tanigawa², Hiroki Sakaji³ (1.Nagaoka University of Technology, 2.IRD Patent Office, 3.The University of Tokyo)

Keywords:NLP, patent analysis, keyword extraction, Graph-based model

Extracting important technical terms representing the technical feature of an invention is useful for patent analysis.
Most previous studies, however, do not consider a structure of an entire patent document, have a problem that semantically miscellaneous keywords are extracted.
In this paper, we propose the graph-based unsupervised important technical term extraction method using the graph representing the structure information of a patent document and its semantic relation.
The performance of the proposed method is compared with previous keyword extraction methods (TF-IDF, TextRank, PositionRank).
The proposed method achieved the highest extraction performance which 65.15 % in F1 score among the comparison methods.
In addition, our method shows the highest extraction performance regardless of the technology field.