Social Influence Dialogue Systems: A Survey of Datasets and Models For Social Influence Tasks

Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 750–766 May 2-6, 2023 ©2023 Association for Computational Linguistics Social Influence Dialogue Systems: A Survey of Datasets and Models For Social Influence Tasks Kushal Chawla1∗ Weiyan Shi2∗ Jingwen Zhang3∗∗ Gale Lucas1∗∗ Zhou Yu2∗∗ Jonathan Gratch1∗∗ 1University of Southern California 2Columbia University 3University of California Davis 1{chawla, lucas, gratch}@ict.usc.edu 2{ws2634, zy2461}@columbia.edu 3jwzzhang@ucdavis.edu Abstract Dialogue systems capable of social influence such as persuasion, negotiation, and therapy, are essential for extending the use of technol- ogy to numerous realistic scenarios. However, existing research primarily focuses on either task-oriented or open-domain scenarios, a cate- gorization that has been inadequate for captur- ing influence skills systematically. There exists no formal definition or category for dialogue systems with these skills and data-driven efforts in this direction are highly limited. In this work, we formally define and introduce the category of social influence dialogue systems that influ- ence users’ cognitive and emotional responses, leading to changes in thoughts, opinions, and behaviors through natural conversations. We present a survey of various tasks, datasets, and methods, compiling the progress across seven diverse domains. We discuss the commonalities and differences between the examined systems, identify limitations, and recommend future di- rections. This study serves as a comprehensive reference for social influence dialogue systems to inspire more dedicated research and discus- sion in this emerging area. 1 Introduction Consider a human user who signs up to interact with a persuasive dialogue system that motivates for engaging in physical exercise. The system: 1) uses social cues like small talk and empathy, useful for providing continued support, and 2) employs persuasive strategies to convince the user who, at least in the short-term, is reluctant to indulge in exercise. Does such a system fit the definition of a task-oriented dialogue system that are traditionally designed to assist users in completing their tasks such as restaurant or flight booking (Zhang et al., 2020c)? Although the system is task-oriented or goal-oriented per se, the task here goes beyond the traditional definition of assisting users, given the ∗Equal contribution, ∗∗Co-supervised project possible misalignment between the goals of the sys- tem and the user. Clearly, this system is also not open-domain (Huang et al., 2020). Although con- versations involve social open-ended interactions, there is still a concrete goal of persuading the user towards a healthier habit. Scenarios similar to above are ubiquitous in ev- eryday life, including games (Peskov et al., 2020), social platforms (Tan et al., 2016), and therapeu- tic interactions (Tanana et al., 2016). Dialogue systems for these applications require a core func- tion in human communication, that is, social in- fluence (Cialdini and Goldstein, 2004; Cialdini, 2009), which involves influencing users’ cogni- tive and emotional responses, leading to changes in thoughts, opinions, and behaviors through natural conversations. This goes beyond what is captured by traditional task definitions in the dialogue com- munity which primarily focus on task completion and social companionship. Despite numerous independent efforts in iden- tifying and …

Content truncated to 500 words