Yahoo! JAPAN R&D

The following papers won the JSAI Annual Conference Award at JSAI 2020 (external link).

News Headline Generation Based on Reinforcement Learning Considering Human Evaluation Yu Yamada (Tokyo Tech), Soichiro Fujita (Tokyo Tech), Tomohide Shibata, Hayato Kobayashi, Hiroaki Taguchi, Manabu Okumura (Tokyo Tech)

This research is done in collaboration with Okumura-Takamura-Funakoshi Group at Tokyo Institute of Technology (external link), and improves the quality of news headline generation by incorporating human evaluation with reinforcement learning. The result of this research is used for the editorial support in Yahoo! JAPAN News. We continue our research to show the effectiveness of NLP in real services.

Extracting interpersonal relationships from Wi-Fi Connection Data Tatsuo Yamashita, Teruhiko Teraoka, Hiroaki Taguchi

This research presents a simple method to detect the occurrence of the communication of mutual employees in the office using only Wi-Fi connection data. First, we collected data of the employee location search system using the Wi-Fi connection data of the devices such as laptop computers, and then we analyzed the time variation in the location information of each device, and lastly, we detected the meeting (time and location sharing) and extracted interpersonal relationships. Results of experiments using actual data suggest that our method works effectively.