Title: Collab-REC: An LLM-based Agentic Framework for Balancing Recommendations in Tourism

Abstract: This paper introduces COLLAB-REC, a novel multi-agent architecture aimed at mitigating popularity bias and enhancing the diversity of tourism recommendations. The system employs three distinct LLM-based agents—specializing in Personalization, Popularity, and Sustainability—to propose city destinations from varied viewpoints. A non-LLM moderator subsequently synthesizes and polishes these suggestions via iterative constrained refinement. This process ensures that every agent’s perspective is incorporated while minimizing redundant or spurious results. Offline evaluations conducted on European city queries, utilizing various LLM sizes and model families, demonstrate that COLLAB-REC outperforms single-agent baselines in both diversity and overall relevance. Notably, the framework successfully highlights less-visited destinations that are typically ignored. By adopting this balanced, context-aware strategy, the system more effectively addresses a wider array of user and systemic concerns, underscoring the value of multi-stakeholder collaboration in LLM-powered recommender systems. The code, data, and associated artifacts can be accessed at https://github.com/ashmibanerjee/collab-rec, with detailed prompts provided in the appendix.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC