cs.IRJul 1, 2025
SaveRethinking Group Recommender Systems in the Era of Generative AI: From One-Shot Recommendations to Conversational Group Decision Support
Organizations: University of Klagenfurt, Austria · University of Sarajevo, Bosnia · Free University of Bozen-Bolzano, Italy
Abstract
More than twenty-five years ago, early ideas emerged on how to design systems that provide recommendations to groups of users rather than individuals. Since then, a wide range of algorithmic approaches has been proposed to elicit individual preferences, aggregate them, and generate group recommendations. However, despite this substantial body of work, relatively few examples of real-world group recommender systems are in use. This observation invites reflection on the assumptions of the academic research, particularly regarding how inter-group communication unfolds and how recommendation-supported decisions should be made. In this essay, we suggest that some of these assumptions and corresponding system designs may not fully align with users' needs or expectations. We therefore outline a forward-looking perspective that explores how insights from group-decision making research and recent advances in Generative AI, including systems such as ChatGPT, could better inform the design of group recommender systems. As one possible direction, we consider settings in which human group members interact and exchange information through a chat based interface, while an AI-based agent supports the decision-making process. Such an approach may contribute to design more natural group and system interactions, and could help facilitate a broader adoption of group recommender systems in practice.
Figures & tables
| Aspect | Insight from the literature | Opportunities with LLMs/GenAI |
|---|---|---|
| Reducing the Choice Set | Group decisions may require several rounds of discussion and comparison. | Support iterative deliberation and adapt recommendations to the evolving decision process. |
| Decision process quality and Choice patterns | The quality of the decision process, including justifiability and fairness, matters alongside the outcome. Group members’ assessment of items and attributes can in fact be influenced by social factors such as conformity or interpersonal relationships. | Identify relevant decision criteria and arguments. Summarize evolving preferences, agreements, disagreements, and establish a shared understanding of alternatives and trade-offs. Encourage consideration of diverse perspectives. |
| Meta-level decision support | Groups may benefit from guidance on how to structure and conduct their decision process. | Suggest suitable decision strategies and provide context-sensitive guidance during discussions. |
Table 1. Selected insights from decision-making research and implications for next-generation GRS.
| Work | Main idea | Opportunities with LLMs/GenAI |
|---|---|---|
| Nguyen et al. ( Nguyen and Ricci, 2017 ) | A chat-based GRS supports dynamic group decision-making by allowing members to discuss and share candidate items, react to them, and obtain aggregated recommendations. | LLMs could continuously interpret the discussion, infer evolving preferences, and provide recommendation support without requiring preferences and reactions to be explicitly encoded through dedicated interface elements. |
| Álvarez-Márquez and Ziegler ( Álvarez Márquez and Ziegler, 2018 ) | An interactive GRS supports negotiation by letting group members first agree on desirable item features and subsequently discuss and evaluate system-generated recommendations. | A tighter integration of conversation and recommendation becomes possible by automatically extracting relevant criteria, preferences, arguments, and changes of opinion from the discussion. |
| Delić et al. ( Delic et al., 2023 ) (CHARM) | A chatbot embedded in an existing messaging platform mediates group decision-making through explicit commands for tasks such as registering suggestions, summarizing preferences, and generating recommendations. | Predefined commands could be replaced by natural-language interaction, with LLMs recognizing user intents and proactively determining when and how to support the decision process. |
Table 2. Selected approaches to chat-based group recommender systems and opportunities enabled by LLMs and Generative AI.
| Work | Challenge / role | Main idea | Opportunities with LLMs/GenAI |
|---|---|---|---|
| BlahBlahBot ( Shin et al., 2021 ) | Orientation / social facilitator | Analyzes participants’ social media content to identify shared interests and suggests conversation topics to initiate interaction. | Shared interests could be inferred from the conversation to help ad-hoc groups start discussing their preferences. |
| IntroBot ( Shin et al., 2023 ) | Orientation / social facilitator | Scaffolds early group interaction through prompts that mark conversation phases, and encourages members who remain inactive. | The early phase of a group decision could be structured, e.g., by eliciting each member’s initial preferences through adaptive prompts. |
| Kim et al. ( Kim et al., 2020 ) | Alignment / facilitator | Monitors participation and proactively encourages less active members to voice their opinion, increasing opinion diversity. | Members whose preferences are not yet known could be detected and asked for their input in a natural, personalized way. |
| DebateBot ( Kim et al., 2021 ) | Alignment / facilitator | Structures deliberative discussions by soliciting opinions from reticent members and asking discussants to share the reasoning behind their proposals. | Context-sensitive prompts could ask members to justify their preferences, supporting more justifiable group decisions. |
| Tilda ∗ ( Zhang and Cranshaw, 2018 ) | Alignment / facilitator | Lets users tag chat messages and provides summaries that help reconstruct long discussions. | Messages could be annotated automatically, e.g., as proposals or preference statements, and the discussion summarized without manual tagging. |
| SolutionChat ∗ ( Lee et al., 2020 ) | Alignment / support for human facilitator | Visualizes the state of the discussion and recommends moderation messages to human moderators. | The evolving discussion could be interpreted to generate moderation actions, potentially taking on parts of the moderator role. |
Table 3. Overview of selected conversational agents for group decision support from the literature, including challenges, agent roles, and opportunities enabled by Generative AI. Challenges and roles follow the categories of ( Yeo et al., 2026 ) ; works marked with ∗ are not part of their review and were assigned by us.
Figure 1. Illustration of an envisioned AI-supported group decision-making scenario in which group members discuss options using a common messaging app, with a group decision-support agent participating in the conversation. The agent’s behavior can be configured for the specific use case, ranging from reactive responses to proactive interventions.
Figure 2. Overview of groups of supported capabilities in the envisioned framework.
| Choice pattern | System Support |
|---|---|
| Attribute-based choice | The system can summarize item features, highlight relevant decision criteria, and explain trade-offs between alternatives. |
| Socially-based choice | The system can present arguments for and against alternatives raised by others and help group members consider each other’s expectations and opinions. |
| Consequence-based choice | The system can help the group explore alternative courses of action and assess their potential consequences (e.g., regarding sustainability). |
| Policy-based choice | The system can help the group select an appropriate decision-making procedure based on the context and guide its implementation. |
| Experience-based choice | The system can recall similar past decisions and help the group assess whether previously used decision strategies are suitable for the current situation. |
| Trial-and-error-based choice | The system can support the exploration of alternatives, potentially through simulations, and help the group evaluate their implications before making a final choice. |
Table 4. Illustrative support mechanisms for different group choice patterns ( Jameson et al., 2022 ) within the envisioned framework.
Figure 3. Illustrative agent-based architecture for implementing the envisioned GRS capabilities, following ( Peng et al., 2025 ) .
More in this topic
Recommender Systems
Recommender systems predict and rank items for people using interaction histories, content, context, and feedback.
See all 521 papers