Open access
Jul 2026
Evaluating reinforcement learning from human feedback for task‐oriented dialogue systems
Experiments on MultiWOZ 2.1 show that the proposed RLHF approach consistently improves corpus‐based evaluation over supervised baselines, while simulator‐based effects remain mixed and backbone‐dependent.
Hyeok‐Min Gwon, Yohan Lee, Jin-Xia Huang et al.
· ETRI Journal · 0 citations