[3Xin4] Poster session 1

[3Xin4-38] Question Generation for Narrowed Retrieval of Patent Incorporating User Intentions

〇Kana Ozaki¹, Yasuhiro Sogawa¹ (1.Research & Development Group, Hitachi, Ltd.)

Keywords:Patent Retrieval, Question Generation, Natural Language Processing

In patent search opeations such as technology trend searches and prior art searches, patent search system is required to obtain the desired set of patent documents with high accuracy. Expert searchers can assemble a search formula appropriately according to their purpose of the search, however novice searchers have difficulty in reflecting their own search intention in the search formula. Therefore, the search results are not often output as intended in a single search. In particularly, if none of the target patent documents appear at the top of the search results, searchers have to look through many patent documents. To tackle this challenge, we propose a question generation method to narrow down the target patent documents from the search results set. In the experiment, we confirmed the effectivenes of our proposed method through the evaluation in patent documents.