[4G1] [General Session] 9. NLP / IR

[4G1-05] A Study on a Dialogue System based on User’s Preference estimated by Input Images

Sae Iijima¹, Kasumi Aoki¹, Katsuyoshi Yamagami^2,3, 〇Ichiro Kobayashi¹, Jun Ozawa² (1. Ochanomizu University, 2. National Institute of Advanced Industrial Science and Technology, 3. Panasonic Corporation)

Keywords:Dialogue, Recipe recommendation, Reinforcement learning

We have developed a dialogue system for recipe recommendation. In particular, we have proposed an idea to allow users to provide the system with a picture of their favorite foods so as they can tell the system their preference on foods without interaction by words. By this, the system can estimate user’s hidden favorites and recommend foods. In the dialogue management, we have optimized dialogue policy using natural actor critic.