[2E2] [General Session] 9. NLP / IR

[2E2-03] Headline Generation of Yahoo! Chiebukuro Based on Learning to Rank

〇Tatsuru Higurashi¹, Hayato Kobayashi^1,2, Kazuma Murao¹, Masuyama Takeshi¹ (1. Yahoo Japan Corporation, 2. RIKEN AIP)

Keywords:Machine Learning, Natural Language Processing, Crowdsourcing

In this paper, we propose a training method for headline generation based on learning-to-rank, so that the appropriate summarization of a question posted by the user is distributed in Yahoo! Chiebukuro.
For headline generation we used data provided from National Institute of Informatics and generated supervised-data through crowdsourcing.
As a result, effective results were obtained compared with the conventional method.