Counselor Reflection Generation with Dynamic Reward Adjustment in Multi-Reward Reinforcement Learning

Bandit-based methods, DynaOpt and C-DynaOpt, outperform existing baselines in enhancing counselor reflection quality.