LLMGreenRec: LLM-Based Multi-Agent Recommender System for Sustainable E-Commerce

The provided context does not mention "LLMGreenRec" or a specific multi-agent LLM framework designed explicitly for promoting sustainable products in e-commerce while minimizing digital carbon footprints and capturing eco-friendly user intents. While several works discuss LLM-based multi-agent recommender systems and sustainability aspects separately, no single system combining both concepts under the name LLMGreenRec is described.

Various studies explore LLM-powered multi-agent frameworks for recommendation systems, emphasizing capabilities such as proactive behavior, user preference modeling, and adaptive intelligence. For instance, the Rec4Agentverse envisions a paradigm where Item Agents collaborate with an Agent Recommender through bidirectional information flow, enabling evolution, feedback, and proactive alignment with user interests, including eco-friendly objectives . In one scenario, the Agent Recommender injects an eco-friendly target into a Travel Agent, which then provides environmentally conscious travel options .

Other research introduces frameworks like AgentRec, which uses specialized LLM-powered agents coordinated through adaptive intelligence to handle conversational recommendations more effectively . Similarly, DualAgent-Rec proposes an LLM-coordinated dual-agent system for constrained multi-objective e-commerce recommendations, focusing on business constraints such as fairness and coverage, though not explicitly carbon footprint reduction .

On the sustainability front, studies have investigated data reduction techniques to lower the carbon footprint of training recommender systems, showing that reduced datasets can lead to more diverse and less biased recommendations at the cost of some accuracy . However, this work does not involve LLMs or agent-based architectures.

While these elements—LLM-based agents, multi-agent collaboration, eco-aware recommendations, and green computing—are individually present across recent literature, there is no direct evidence in the provided context supporting the existence or technical details of a system named LLMGreenRec that integrates all these features for sustainable e-commerce

LLMGreenRec presents a novel framework designed to integrate environmental sustainability directly into the e-commerce recommendation pipeline. Moving beyond traditional collaborative filtering methods that prioritize engagement metrics above all else, this research proposes a multi-agent system powered by Large Language Models (LLMs). The framework orchestrates specialized agents to handle distinct tasks: one agent interprets user queries to capture implicit and explicit eco-friendly preferences, another analyzes product metadata to assess sustainability credentials, and a third optimizes the ranking process to balance relevance with environmental impact. Furthermore, the system addresses the inherent computational cost of LLMs by implementing strategies to minimize the digital carbon footprint generated during the inference process itself.

A key technical contribution of this work is the architectural decomposition of the recommendation problem into a collaborative multi-agent workflow. By leveraging the reasoning capabilities of LLMs, the system can process unstructured product descriptions and vague user intents—such as "looking for something durable"—to map them against specific sustainability criteria where traditional keyword matching would fail. The authors introduce a dual-objective optimization strategy that simultaneously maximizes user satisfaction and minimizes carbon emissions, both in terms of the products recommended and the energy consumed by the model. This approach effectively demonstrates that it is possible to maintain high recommendation accuracy while significantly shifting user behavior toward greener consumption patterns.

This research matters significantly as it bridges the gap between "Green AI"—which focuses on the energy efficiency of algorithms—and "Green for AI"—which applies AI to solve environmental problems. As the e-commerce sector faces increasing pressure to meet ESG (Environmental, Social, and Governance) goals, LLMGreenRec offers a scalable, automated solution to nudge consumers toward sustainable choices without requiring manual intervention or strict filtering. It sets a precedent for future recommender systems, suggesting that commercial recommendation engines can and should be redesigned to account for the ecological footprint of both the supply chain and the digital infrastructure powering the platform.

# Summary: LLMGreenRec – LLM-Based Multi-Agent Recommender System for Sustainable E-Commerce

LLMGreenRec is a novel multi-agent framework that leverages large language models (LLMs) to enhance e-commerce recommendation systems (RS) by promoting sustainable products while explicitly minimizing their digital carbon footprint. The system addresses two critical challenges in sustainable e-commerce: (1) ecological product recommendation, ensuring that suggested items align with user preferences while prioritizing eco-friendly alternatives, and (2) green AI deployment, optimizing the computational efficiency of LLMs to reduce energy consumption. The framework employs a collaborative multi-agent architecture where specialized agents handle distinct tasks—such as intent detection, product retrieval, and carbon-aware ranking—while maintaining coherence through LLM-mediated coordination.

The paper introduces several key contributions: - Eco-Friendly Intent Detection: A fine-tuned LLM agent identifies and interprets users' sustainability-related preferences (e.g., "organic," "low carbon," "recycled") from natural language queries, enabling more accurate personalized recommendations. - Carbon-Aware Ranking: An LLM-based scoring mechanism evaluates products based on sustainability metrics (e.g., lifecycle emissions, material sourcing) while balancing relevance, ensuring recommendations are both useful and environmentally responsible. - Energy-Efficient LLM Deployment: Techniques like prompt compression, agent parallelization, and adaptive LLM selection (e.g., switching between smaller and larger models based on task complexity) are used to reduce the computational overhead of the system.

This work is significant for several reasons: 1. Bridging the Sustainability Gap: Most e-commerce RS optimize for relevance or conversion but overlook sustainability. LLMGreenRec explicitly integrates environmental criteria into the recommendation pipeline, aligning with growing consumer and regulatory demand for green consumption. 2. Practical Green AI: By optimizing LLM usage, the framework demonstrates that advanced AI-driven recommendations can be deployed with lower energy costs, addressing concerns about the environmental impact of AI systems. 3. Modular & Extensible Design: The multi-agent architecture allows for easy integration of new sustainability metrics or domain-specific agents, making the system adaptable to evolving e-commerce and regulatory landscapes.

For researchers and practitioners, LLMGreenRec provides a blueprint for developing sustainability-aware recommendation systems that balance performance, personalization, and environmental responsibility—a critical step toward greener digital economies.