**MACRec: A Multi-Agent Collaboration Framework for Recommendation** **Authors:** Zhefan Wang **Abstract:** This paper introduces MACRec, a novel framework designed to enhance recommendation systems through multi-agent collaboration. Unlike existing work that focuses on using agents for user/item simulation, MACRec aims to deploy multi-agents directly to tackle recommendation tasks. The framework includes specialized agents such as Manager, User/Item Analyst, Reflector, Searcher, and Task Interpreter, each with distinct roles and workflows. MACRec addresses various recommendation tasks, including rating prediction, sequential recommendation, conversational recommendation, and explanation generation. The framework is publicly available, and an online web interface visualizes the collaborative process. **Key Contributions:** 1. **New Multi-agent Collaboration Framework:** MACRec is the first framework to directly use multi-agents for recommendation tasks, differencing it from previous studies that focus on agent-based simulation. 2. **Diverse Applications:** MACRec supports multiple recommendation scenarios, including rating prediction, sequential recommendation, explanation generation, and conversational recommendation. 3. **User-friendly Online Web Interface:** An online interface visualizes the collaborative process, making it accessible and user-friendly. **Introduction:** Recommender systems (RSs) are crucial for improving user experience and platform benefits. LLM-based agents have shown promise in handling complex tasks, but their integration into RSs has been limited. MACRec addresses this gap by leveraging the diverse capabilities of multi-agents to tackle recommendation tasks more effectively. **Related Work:** - **Agent-based Recommendation:** Research on agent-based recommendation can be categorized into simulation-oriented and recommender-oriented approaches. MACRec differs from these by focusing on direct agent collaboration for recommendation tasks. - **Multi-agent Collaboration:** Multi-agent systems have evolved with foundational concepts of coordination and communication. MACRec builds on these concepts to achieve better performance in complex tasks. **MACRec Framework:** - **Framework Overview:** The framework involves a Task Interpreter, Manager, Reflector, User/Item Analyst, Searcher, and Task Interpreter working collaboratively to solve recommendation tasks. - **Agent Roles:** - **Manager:** Oversees the collaboration and manages sub-tasks. - **Reflector:** Analyzes the correctness of the Manager's answers and suggests improvements. - **User/Item Analyst:** Examines user and item characteristics. - **Searcher:** Retrieves additional information using search tools. - **Task Interpreter:** Translates dialogues into executable tasks. **Applications:** - **Rating Prediction:** Analyzes user preferences and item characteristics. - **Sequential Recommendation:** Models long-term and short-term user interests. - **Explanation Generation:** Provides detailed explanations for recommendations. - **Conversational Recommendation:** Engages users in dialogues to refine preferences. **Conclusion:** MACRec is a novel LLM-based multi-agent collaboration framework for recommendation systems. It directly addresses recommendation tasks through the collaborative efforts of specialized agents, offering applications in various recommendation**MACRec: A Multi-Agent Collaboration Framework for Recommendation** **Authors:** Zhefan Wang **Abstract:** This paper introduces MACRec, a novel framework designed to enhance recommendation systems through multi-agent collaboration. Unlike existing work that focuses on using agents for user/item simulation, MACRec aims to deploy multi-agents directly to tackle recommendation tasks. The framework includes specialized agents such as Manager, User/Item Analyst, Reflector, Searcher, and Task Interpreter, each with distinct roles and workflows. MACRec addresses various recommendation tasks, including rating prediction, sequential recommendation, conversational recommendation, and explanation generation. The framework is publicly available, and an online web interface visualizes the collaborative process. **Key Contributions:** 1. **New Multi-agent Collaboration Framework:** MACRec is the first framework to directly use multi-agents for recommendation tasks, differencing it from previous studies that focus on agent-based simulation. 2. **Diverse Applications:** MACRec supports multiple recommendation scenarios, including rating prediction, sequential recommendation, explanation generation, and conversational recommendation. 3. **User-friendly Online Web Interface:** An online interface visualizes the collaborative process, making it accessible and user-friendly. **Introduction:** Recommender systems (RSs) are crucial for improving user experience and platform benefits. LLM-based agents have shown promise in handling complex tasks, but their integration into RSs has been limited. MACRec addresses this gap by leveraging the diverse capabilities of multi-agents to tackle recommendation tasks more effectively. **Related Work:** - **Agent-based Recommendation:** Research on agent-based recommendation can be categorized into simulation-oriented and recommender-oriented approaches. MACRec differs from these by focusing on direct agent collaboration for recommendation tasks. - **Multi-agent Collaboration:** Multi-agent systems have evolved with foundational concepts of coordination and communication. MACRec builds on these concepts to achieve better performance in complex tasks. **MACRec Framework:** - **Framework Overview:** The framework involves a Task Interpreter, Manager, Reflector, User/Item Analyst, Searcher, and Task Interpreter working collaboratively to solve recommendation tasks. - **Agent Roles:** - **Manager:** Oversees the collaboration and manages sub-tasks. - **Reflector:** Analyzes the correctness of the Manager's answers and suggests improvements. - **User/Item Analyst:** Examines user and item characteristics. - **Searcher:** Retrieves additional information using search tools. - **Task Interpreter:** Translates dialogues into executable tasks. **Applications:** - **Rating Prediction:** Analyzes user preferences and item characteristics. - **Sequential Recommendation:** Models long-term and short-term user interests. - **Explanation Generation:** Provides detailed explanations for recommendations. - **Conversational Recommendation:** Engages users in dialogues to refine preferences. **Conclusion:** MACRec is a novel LLM-based multi-agent collaboration framework for recommendation systems. It directly addresses recommendation tasks through the collaborative efforts of specialized agents, offering applications in various recommendation