1 paper · 1 filter
Chenghui Zou, Ning Wang, Tiesunlong Shen +5
Large language models (LLMs) have advanced emotional support conversation, but existing alignment methods rely mainly on sparse preferences at the response level or outcomes at the…