Abstract: Generative AI's rapid evolution has made dialogue systems indispensable tools. While persuasive strategies have been incorporated in dialogue systems to provide personalized services, current research primarily focuses on studying persuasive strategies from persuader’s perspective, with limited exploration of persuadee's susceptibility towards these strategies. To bridge this gap, we introduce a novel task called Susceptibility Strategy Detection, aimed at identifying the persuasive strategies that users are most susceptible to. To support this new task, we develop a refined dataset P4G+, and propose a dual attitude-sensitive framework to detect susceptibility strategy by analyzing the persuasive process, user interactions, and content within dialogues. Comprehensive experiments have demonstrated the efficacy of our approach in identifying users’ susceptible strategies. The code and dataset will be made available upon acceptance of this paper.
Paper Type: Long
Research Area: Dialogue and Interactive Systems
Research Area Keywords: conversational modeling
Contribution Types: NLP engineering experiment, Data resources
Languages Studied: English
Submission Number: 5651
Loading