Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Entity Linking using LLMs for Automated Product Carbon Footprint Estimation

Steffen Castle, Julian Moreno Schneider, Leonhard Hennig, Georg Rehm

TL;DR

The paper tackles the problem of automating product carbon footprint estimation by linking component-level BOM data to LCA database entries. It introduces a three-module pipeline that leverages a pretrained LLM to generate detailed component-context descriptions, augments this with relevant datasheet content, and uses semantic similarity in a FAISS-based vector store to map to LCA entries (primarily ecoinvent). Empirical results on a limited dataset show that combining LLMs with datasheet context can match or exceed non-expert human performance for top recommendations, while semantic similarity alone underperforms, highlighting the value of language-model context in fine-grained entity linking. The approach promises scalable, less-manual carbon-footprint estimation, with practical impact for industry adoption and regulatory reporting, and points to future work on broader evaluation and web-context integration.

Abstract

Growing concerns about climate change and sustainability are driving manufacturers to take significant steps toward reducing their carbon footprints. For these manufacturers, a first step towards this goal is to identify the environmental impact of the individual components of their products. We propose a system leveraging large language models (LLMs) to automatically map components from manufacturer Bills of Materials (BOMs) to Life Cycle Assessment (LCA) database entries by using LLMs to expand on available component information. Our approach reduces the need for manual data processing, paving the way for more accessible sustainability practices.

Entity Linking using LLMs for Automated Product Carbon Footprint Estimation

TL;DR

The paper tackles the problem of automating product carbon footprint estimation by linking component-level BOM data to LCA database entries. It introduces a three-module pipeline that leverages a pretrained LLM to generate detailed component-context descriptions, augments this with relevant datasheet content, and uses semantic similarity in a FAISS-based vector store to map to LCA entries (primarily ecoinvent). Empirical results on a limited dataset show that combining LLMs with datasheet context can match or exceed non-expert human performance for top recommendations, while semantic similarity alone underperforms, highlighting the value of language-model context in fine-grained entity linking. The approach promises scalable, less-manual carbon-footprint estimation, with practical impact for industry adoption and regulatory reporting, and points to future work on broader evaluation and web-context integration.

Abstract

Growing concerns about climate change and sustainability are driving manufacturers to take significant steps toward reducing their carbon footprints. For these manufacturers, a first step towards this goal is to identify the environmental impact of the individual components of their products. We propose a system leveraging large language models (LLMs) to automatically map components from manufacturer Bills of Materials (BOMs) to Life Cycle Assessment (LCA) database entries by using LLMs to expand on available component information. Our approach reduces the need for manual data processing, paving the way for more accessible sustainability practices.

Paper Structure

This paper contains 12 sections, 5 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. Background
  4. LCA Database
  5. Bills of Materials
  6. Component Datasheets
  7. Methodology
  8. Datasheet Selection
  9. LLM Querying
  10. Semantic Similarity Matching
  11. Results and Discussion
  12. Conclusion and Future Work

Figures (5)

  • Figure 1: An illustrative example of the process name and description from a LCA database.
  • Figure 2: An excerpt from a BOM. Note that material codes are often ambiguous or obscure, requiring specialized knowledge to correctly identify.
  • Figure 3: Architecture of the proposed pipeline, made up of three modules: Document retrieval, LLM querying, and database ranking.
  • Figure 4: Sample LLM response for a BOM component. Inclusion of datasheet context appears to considerably improve reliability of responses.
  • Figure 5: Results of evaluation. Our proposed approach is able to match the performance of a human on the top 5 matches, and exceeds human performance on the top match.