[2M2] AI応用-産業応用(3)

[2M2-01] Utilizing deception information for dialog management of doctor-patient conversations.

〇Tung The Nguyen¹, Koichiro Yoshino¹, Sakriani Sakti¹, Satoshi Nakamura¹ (1. Graduate School of Information Science, Nara Institute of Science and Technology)

キーワード：POMDP, Reinforcement Learning, Multi-modal

Almost all of existing negotiation systems assume that their interlocutors (the user) are telling the truth. However, in negotiations, participants can tell lies to earn a profit. In this research, we proposed a negotiation dialog management system that detects user's lies and designed a dialog behavior on how should the system react with. As a typical case, we built a dialog model of doctor-patient conversation on living habits domain. We showed that we can use partially observable Markov decision process (POMDP) to model this conversation and use reinforcement learning to train the system's policy.