MACS: Hybrid Multi-Agent Framework for Reliable Conversational E-Commerce Recommendations
A recent study presents MACS (Multi-Agent Commerce System), a hybrid multi-agent framework aimed at improving the reliability of conversational recommendation systems in e-commerce, specifically within fixed-catalog environments. The research, which can be found on arXiv (2608.14068), tackles the issue of generating recommendations solely from a merchant's catalog, without relying on web searches or unsupported assertions. MACS utilizes large language models (LLMs) for tasks involving language, such as understanding user inquiries, gathering preferences, and crafting responses. Meanwhile, crucial operations like product retrieval and brand exclusion are performed deterministically. This approach ensures user needs are met, aligns with available inventory, and maintains preferences throughout multiple conversational exchanges, emphasizing the significance of reliability in AI-driven commerce.
Key facts
- MACS is a hybrid multi-agent framework for conversational recommendation in fixed-catalog settings.
- It uses LLMs for language-facing tasks and deterministic methods for correctness-critical operations.
- The framework ensures recommendations are drawn only from a merchant's fixed catalog.
- It addresses challenges of reliability under hard constraints in e-commerce.
- The paper is available on arXiv with identifier 2608.14068.
- MACS aims to satisfy user requirements, remain grounded in inventory, and preserve preferences across turns.
- It includes operations like product retrieval, hard-constraint filtering, brand exclusion, and progressive relaxation.
Entities
Institutions
- arXiv