FADO: Feedback-Aware Double COntrolling Network for Emotional Support Conversation

Emotional Support Conversation (ESConv) aims to reduce help-seekers’ emotional distress with a supportive strategy and response. It is essential for the supporter to select an appropriate strategy according to the feedback of the help-seeker (e.g., emotion change during dialog turns, etc.) in ESConv. However, previous methods mainly rely on the dialog history to select the strategy and ignore the help-seeker’s feedback, causing wrong and user-irrelevant strategy predictions. Meanwhile, these methods only model the context-to-strategy flow but pay less attention to the strategy-to-context flow involving the strategy-related context for generating strategy-constrained responses. In this paper, a Feedback-Aware Double COntrolling Network (FADO) is proposed to make a strategy schedule and generate supportive responses. The core modules in FADO include a dual-level feedback strategy selector and a double control reader, where the former leverages the turn-level and conversation-level feedback to encourage or penalize strategies, and the latter constructs a novel strategy-to-context flow to generate strategy-constrain responses. Besides, a strategy dictionary is designed to enrich the semantic information of the strategy and improve the quality of the strategy-constrained response. Experimental results on ESConv indicate that the proposed FADO achieves SOTA performance in terms of strategy selection and response generation. Our code is available at https://github.com/Thedatababbler/FADO.