Estimating Product Carbon Footprint via Large Language Models for Sustainable Recommender Systems

The environmental impact of e-commerce continues to grow, driven by global supply chains and unsustainable consumption patterns. In this context, accurately estimating product carbon footprint (PCF) is a fundamental step to making decisions that are more environmentally conscious. However, this process is complex to implement due to the lack of comprehensive information about the carbon footprint of items and production processes. Accordingly, this study introduces a novel methodology that leverages Large Language Models (LLMs) to estimate the life cycle carbon footprint (CO2) of commercial products using unstructured textual data, such as product descriptions and metadata. The approach enables the automatic augmentation of product datasets with environmental indicators. To demonstrate the practical relevance of this framework, we integrate the CO2 information into a recommender system, enabling the generation of personalized yet more environmentally sustainable suggestions. Experimental results on an Amazon Electronics dataset confirm the effectiveness of LLMs in approximating emission values, offering a reliable strategy to support both research and policy efforts targeting Goal 12 of the UN Sustainable Development Goals: Responsible Consumption and Production.