ComPeer: A Generative Conversational Agent for Proactive Peer Support

TL;DR

This work tackles the challenge of sustaining peer support in mental health contexts by introducing ComPeer, a proactive generative conversational agent built on large language models. ComPeer integrates memory, event detection, reflection, and scheduling to proactively initiate peer-support dialogues tailored to a user’s state and persona, moving beyond traditional user-initiated or rule-based systems. In a one-week randomized study with 24 university students, ComPeer demonstrated comparable stress relief to a baseline CA while delivering higher engagement and perceived quality of advice, with qualitative data highlighting both benefits and concerns of proactive, AI-driven care. The study provides design principles and practical considerations for deploying proactive generative agents in healthcare scenarios, and discusses generalizability, safety, and ethical implications for future work. Overall, ComPeer advances proactive, adaptive peer support through an interpretable architecture and empirical user data, offering a blueprint for deploying ethical, user-centered AI in non-clinical mental health support.

Abstract

Conversational Agents (CAs) acting as peer supporters have been widely studied and demonstrated beneficial for people's mental health. However, previous peer support CAs either are user-initiated or follow predefined rules to initiate the conversations, which may discourage users to engage and build relationships with the CAs for long-term benefits. In this paper, we develop ComPeer, a generative CA that can proactively offer adaptive peer support to users. ComPeer leverages large language models to detect and reflect significant events in the dialogue, enabling it to strategically plan the timing and content of proactive care. In addition, ComPeer incorporates peer support strategies, conversation history, and its persona into the generative messages. Our one-week between-subjects study (N=24) demonstrates ComPeer's strength in providing peer support over time and boosting users' engagement compared to a baseline user-initiated CA.

Figures (7)

• Figure 1: Architecture and an example user scenario of ComPeer, a generative agent for proactive peer support.
• Figure 2: Procedure of our one-week between-subjects study (ComPeer vs baseline conversational agent (CA)) and extended one-week study in which all participants converse with ComPeer.
• Figure 3: (a) Changes of participants’ perceived stress measured by PSS-10 over the user study. (b) The regression result of Q1.
• Figure 4: Results of each topic. A is the average satisfaction for each proactive dialogue, B is the average satisfaction between the two major topics, and C is the satisfaction across each sub-topic. Note: + : .05 < $p$ < .10.
• Figure 5: (a) The average rounds of conversation in each group, "C" represents the ComPeer group, and "B" represents the baseline group. (b) The average rounds of each topic in two groups. Note: + : .05 < $p$ < .10, * : $p$ < .05, **: $p$ < .01.