UX Research on Conversational Human-AI Interaction: A Literature Review of the ACM Digital Library

Abstract: Early conversational agents (CAs) focused on dyadic human-AI interaction between humans and the CAs, followed by the increasing popularity of polyadic human-AI interaction, in which CAs are designed to mediate human-human interactions. CAs for polyadic interactions are unique because they encompass hybrid social interactions, i.e., human-CA, human-to-human, and human-to-group behaviors. However, research on polyadic CAs is scattered across different fields, making it challenging to identify, compare, and accumulate existing knowledge. To promote the future design of CA systems, we conducted a literature review of ACM publications and identified a set of works that conducted UX (user experience) research. We qualitatively synthesized the effects of polyadic CAs into four aspects of human-human interactions, i.e., communication, engagement, connection, and relationship maintenance. Through a mixed-method analysis of the selected polyadic and dyadic CA studies, we developed a suite of evaluation measurements on the effects. Our findings show that designing with social boundaries, such as privacy, disclosure, and identification, is crucial for ethical polyadic CAs. Future research should also advance usability testing methods and trust-building guidelines for conversational AI.

• The paper systematically differentiates between polyadic and dyadic conversational agent research, highlighting the limited number of polyadic studies in the ACM Digital Library.
• The paper employs mixed-method approaches, including qualitative thematic analysis and topic modeling, to demonstrate group-level benefits like improved communication efficiency and balanced participation.
• The paper underscores the need for boundary-aware, privacy- and ethics-sensitive CA designs to address challenges in human-human interaction and support effective group dynamics.

Literature Synthesis of UX Research for Polyadic and Dyadic Conversational Agents

Introduction and Rationale

Conversational agents (CAs), encompassing chatbots, intelligent assistants, and embodied agents, have undergone a significant evolution from dyadic (one-on-one) to polyadic (multiparty) interaction paradigms. This paper presents a comprehensive literature review of user experience (UX) research on conversational human-AI interaction, focusing particularly on polyadic CAs—systems that mediate or facilitate interactions between multiple human users. The review draws from 1,302 papers in the ACM Digital Library, ultimately analyzing 36 polyadic and 135 dyadic UX evaluation studies. The review specifically addresses (1) human-human interaction challenges addressed by CAs, (2) major research interests and distinctions between dyadic and polyadic work, (3) predominant design/evaluation practices, (4) empirically supported effects of polyadic CAs on group interaction, (5) common UX evaluation metrics, and (6) critical but overlooked design issues, especially pertaining to boundaries, ethics, and privacy.

Fundamental Human-Human Interaction Challenges in CA Design

The analysis identifies four principal classes of challenges in human-human collaboration that polyadic CAs aim to address:

• Inefficient Communication: Unstructured group communications, difficulties in consensus-reaching, conversational flow management, and coordination overhead motivate CA-mediated workflow and discourse structuring features.
• Lack of Engagement: Participation imbalance, engagement drop-off, and difficulties in stimulating productive peer discussion underpin CA interventions that prompt contributions, balance floor time, and inject prompts or ice-breakers.
• Barriers in Relational Maintenance: Lack of emotional awareness, trust-building, and emotion regulation in distributed or virtual teams leads to exploration of CAs as mediators for sentiment analysis and supportive feedback.
• Need for Building Connections: CAs are deployed to facilitate grounding, establish commonality, and mitigate cultural or social cold-starts in both dyadic and polyadic interactions.

Research Scope, Methods, and Topics

The review employs a mixed-method approach combining qualitative thematic analysis and topic modeling, parsing contrasts between dyadic and polyadic works. The investigation reveals that polyadic CA research is nascent and fragmented, with significantly fewer publications and theoretical models than dyadic work. Most impactful studies in the polyadic domain cluster around educational technology, collaborative work, and large-scale online communities.

Figure 2: Comparison of topical emphases in polyadic and dyadic CA papers, illustrating a higher focus on group-level phenomena and social behavior in polyadic studies.

As shown in Figure 2, the dominant polyadic topics pertain to group discussion, social behavior, education, and embodiment, while dyadic literature is concentrated around user-agent exchanges and conversational design. Topic allocations highlight the distinct research questions each interaction paradigm raises and the limited cross-pollination of frameworks.

Polyadic CA Design Practices and Empirically Evaluated Effects

Shared and unique aspects of polyadic versus dyadic CA design are catalogued:

• Application Domains: Polyadic CAs are prevalent in educational settings, online communities, productivity/work, and larger virtual environments.
• Interaction Modality: Most polyadic CA systems support text-based or multimodal input; embodiment is more common in group contexts requiring visible social signaling.
• Evaluation Methods: Experiments, surveys, interviews, log analysis, and Wizard-of-Oz are widely represented. Survey and log-based analyses are used to quantify communication effectiveness, engagement, and satisfaction.
• Unique Practices: Polyadic CAs introduce considerations of relationship type (co-learners, co-workers, community members) and social scale (small teams to large communities). Design methods are often less theory-driven and lack systematic empiricism found in dyadic CA work.

Robust empirical evidence demonstrates the following group-level affordances of polyadic CAs:

• Enhanced Communication Efficiency: Tangible improvements in consensus-reaching, topic comprehension, and workflow management via CA mediation.
• Increased Engagement and Participation Balance: CARE interventions foster more even participation and facilitate engagement of quieter members.
• Better Relational Maintenance: Positive impacts on group trust, emotional regulation, and social support, particularly via targeted emotional feedback and trust-building behaviors.
• Facilitation of Connection-Building: CAs can lower barriers to first-contact, especially in cross-cultural or unfamiliar group settings, though nuanced biases and perceptions emerge depending on design.

Overlooked Issues: Social Boundaries, Privacy, and Ethics

The review finds several recurring and underexamined design challenges:

• Boundary Awareness: Essential to mitigate inappropriate CA interventions and manage user attention. Polyadic CAs must strike a balance between salience (“visibility”) and non-intrusiveness (“ignorable” design).
• Accountability and Role Clarity: Ambiguity in CA ownership, alignment, and neutrality in conflicts can aggravate privacy concerns and affect trust.
• Privacy Management: There is insufficient attention to how CAs should negotiate the disclosure, control, and turbulence of private/shared group information, especially in sensitive or high-stakes environments.
• Ethical Design: Issues such as invisible influence on group deliberation, bias propagation, and participant protection (especially for marginalized groups) remain partially addressed.

The notion of boundary-aware design is advanced, advocating for CAs that are sensitive to human-to-human and human-to-agent social boundaries—grounded in theories such as Communication Privacy Management and complemented by practical HCI guidelines for transparency, user control, and minimal invasiveness.

Theoretical and Practical Implications

From a theoretical standpoint, the field lacks a unified framework for polyadic CA design and evaluation akin to those supporting dyadic interaction. The review recommends infusion of relevant social psychological, sociological, and communication theories (e.g., balance theory, self-extension, CASA) to rationalize and structure CA interventions in multiparty settings.

Practically, the research highlights an urgent need for standardized metrics for group-level conversational outcomes, richer empirical studies in diverse application domains (especially public services and health), and codification of design guidelines that reflect the unique complexities of polyadic interaction.

Future developments in communicative AI will likely center on:

• Expansion of relational attributions and social roles in CA-mediated group interaction.
• Methodologically rigorous evaluation of CAs as community members and workflow mediators beyond dyadic discourse.
• Integration of privacy- and ethics-aware design patterns, especially for boundary management in dynamically evolving groups.

Conclusion

This literature review establishes foundational knowledge regarding the state and challenges of UX research on polyadic conversational agents within the ACM research corpus. It systematically delineates design practices, evaluation trends, empirically evidenced effects, and key open issues unique to polyadic interaction, culminating in a call for boundary, privacy, and ethics-aware CA design. The practical and theoretical insights presented signal critical directions for the maturation of conversational human-AI interaction at scale.