Brief

Researchers have developed a method to estimate product carbon footprints for e-commerce recommendations, even when labels are missing. This is achieved by inferring carbon data using semantic similarity and LLM prompting. A post-hoc re-ranking strategy then balances predicted user engagement with estimated carbon impact, demonstrating significant carbon reductions with minimal loss in engagement across various product categories. AI

Researchers have developed a method to estimate the carbon footprint of e-commerce products, even when labels are missing. This is achieved by using LLMs and semantic similarity to infer carbon footprints from a small set of assessed products. The system then re-ranks product recommendations to balance user engagement with carbon reduction, demonstrating that significant carbon savings are possible with minimal impact on user interest. AI