Joint Semantic and Strategy Matching for Persuasive Dialogue
Chuhao Jin, Yutao Zhu, Lingzhen Kong, Shijie Li, Xiao Zhang, Ruihua Song, Chen Xu, Chen Huan, Yuchong Sun, Yu Chen, Jun Xu · 2023
Persuasive dialogue aims to persuade users to achieve specific goals through conversations.While previous models have achieved notable successes, they mostly rely on matching utterance semantics and neglect an important aspect: the strategies of a conversation, such as emotional-appeal and foot-in-door.In contrast to utterance semantics, conversation strategies are high-level concepts, which can be informative and provide complementary information, contributing to more effective persuasion.This paper proposes a novel persuasion framework that combines the modeling of conversation semantics and strategies.To accomplish this objective, we design a BERT-like module and an auto-regressive predictor that match the semantics and strategies, respectively.Experimental results indicate that our proposed approach can significantly improve the state-of-the-art baseline by 5% on a small dataset and 37% on a large real-world dataset in terms of [email protected] online evaluation shows that our approach improves the ultimate goal of persuasion in real-world applications.