Rethinking Group Recommender Systems in the Era of Generative AI: From One-Shot Recommendations to Agentic Group Decision Support
Source: arXiv:2507.00535 — July 2025
Authors: Dietmar Jannach, Amra Delić, Francesco Ricci, Markus Zanker
Overview
This position paper from the recommender systems research community argues for a fundamental rethinking of group recommender systems (GRS) in the era of generative AI and large language models (LLMs).
Key points
- From one-shot lists to decision support: Traditional GRS aggregate individual preferences and output a single ranked list of items for a group. The authors propose shifting toward *agentic group decision support*, where conversational agents actively help groups deliberate, negotiate, and converge on decisions.
- LLMs as group facilitators: Generative AI enables multi-turn natural language interaction, contextual preference understanding, and multi-hop reasoning — capabilities that allow systems to participate in (or mediate) the group decision process rather than only predict a final choice.
- Broadened scope: The relevant problem space extends beyond item ranking to open-domain information access, conversational search, semantic understanding of group goals, and end-to-end architectures that combine external knowledge sources with generative models.
- Research agenda: The paper outlines a transition path covering group preference elicitation, LLM-based facilitation of consensus building, and the design of agentic pipelines (retrievers, planners, generators, feedback mechanisms).
- Open challenges: Evaluation credibility for group settings, latency and cost constraints, hallucination and safety risks, cross-lingual and multimodal extension, and fairness/conflict handling among group members.
- 360Brew: A Decoder-only Foundation Model for Personalized Ranking
- Actions Speak Louder than Words: Trillion-Parameter Sequential Transducers
- Bridging Language and Items for Retrieval and Recommendation
- Data-efficient Fine-tuning for LLM-based Recommendation (SIGIR 2024)
Implications for search and recommendation
1. Architecture: Cascade retrieve-rerank-generate remains mainstream, but agentic paradigms make retrieval strategy itself a learnable, sequential decision process. 2. Evaluation: Static offline metrics (e.g., nDCG) increasingly diverge from real group satisfaction; task success and process quality metrics matter more in interactive group settings. 3. Deployment: Latency, cost, explainability, and safety remain hard constraints for production systems; academic benchmarks alone are insufficient.