Enhancing a Conversational Agent with Social Cues: An Investigation into the Impact of Shared Identity and Goal Setting on ‘BLISS’

Enhancing a Conversational Agent with Social Cues: An Investigation into the Impact of Shared Identity and Goal Setting on ‘BLISS’

Submission for CLIN33 Conference

Nikki Evers Iris Hendrickx

Radboud University Radboud University

nikki.evers@ru.nl iris.hendrickx@ru.nl

Abstract

This research investigates the impact of social cues, specifically shared identity and goal setting, on self-disclosure and user experience in the conversational agent BLISS. BLISS is a spoken dialogue conversational agent collecting data about health and well-being, designed to aid in self-management of health, wellness, and psychological well-being. A comprehensive literature review has been conducted in the field of communication theory to identify and explore the relevance of these social cues, which have been under-researched in the context of conversational agents. The choice of shared identity and goal setting as the focus of this study is based on research in human-human conversations, which has demonstrated the relevance and potential impact of these social cues. Shared identity contributes to the establishment of common ground and fosters a sense of relational closeness, often making the conversation feel more like a conversation with a friend rather than with a stranger. Additionally, clearly defining a goal in the conversation can shift the social goal away from presenting a socially acceptable image of oneself, to other social goals like intimacy, relief, control, or identity clarification, decreasing the risk of disclosing more information. Research has shown that both shared identity and goal-setting can increase disclosure and trust. Building on this knowledge, these social cues have been implemented into the BLISS agent to create more natural and engaging conversations, enhance trust and user experience, and elicit increased self-disclosure.

The research adopts a 2x2 experimental design, randomly assigning participants to one of four groups: shared identity, goal setting, shared identity and goal setting, and control. The study aims to recruit 120 Dutch-speaking individuals aged 18 and above, specifically targeting two communities, bee holders and members of a specific tennis club, to establish shared identities during conversations in the shared identity conditions. Data collection will involve participants engaging in one conversation with the BLISS agent, followed by the completion of a questionnaire to assess user experience and relational perceptions. Self-disclosure will be analyzed based on the breadth, duration, and depth of disclosed information. Additionally, self-report measures will assess satisfaction, social attraction, relational perception, trust, empathy, perceived similarity, and perceived conversation goal.

This project aims to contribute to the field of computational linguistics by providing insights into the application of social dynamics to conversational agents, contributing to the better understanding of user experiences. The goal is to improve the understanding of how social cues, shared identity and goal setting, impact user experience, perception of conversational agents and disclosure. The project aligns with the broader goals of the BLISS project, aiming to support individuals in managing their mental well-being. By implementing the novel approach of using shared identity and goal-setting, we aim to enhance the user experience, in turn improving user-agent relationships, and thereby improving the support provided by the BLISS agent.

Keywords: Conversational Agents, Social Cues, Shared Identity, Goal Setting,

Self-Disclosure, User Experience.