Открытый домен многопроцессорного разговора

AbstractTo address the challenge of policy learning in open-domain multi-turn conversation, we propose to represent prior information about dialog transitions as a graph and learn a graph grounded dialog policy, …...