MASCOT: Towards Multi-Agent Socio-Collaborative Companion Systems
Researchers have developed MASCOT, a new framework designed to improve multi-agent systems for collaborative companion roles. MASCOT addresses issues like persona collapse and redundant dialogue by employing a bi-level optimization strategy. This approach includes fine-tuning individual agents for distinct identities and promoting diverse, productive group discourse. Evaluations show MASCOT significantly enhances persona consistency and social contribution in multi-agent interactions. AI
IMPACT Enhances multi-agent systems by improving persona consistency and dialogue diversity, potentially leading to more sophisticated AI companions.