Embedding models compared for environmental social governance text tasks
Benchmarking Embedding Models for ESG Data
Information Retrieval
Summary
Environmental, Social, and Governance (ESG) information helps companies be responsible and clear about their impact. The authors studied how well different computer models turn ESG words into numbers that computers can understand better. They created a special set of ESG examples to test fourteen different models to see which ones perform best at finding relevant information and generating answers using that data. Their tests showed that some models work better than others, with those based on Qwen3 doing the best overall. This helps people who build tools for ESG understanding pick better models.
What this means in practice
- •For financial data teams: Improve retrieval and generation of ESG insights by choosing top-performing embedding models for processing ESG text data.
- •For corporate sustainability teams: Enhance internal ESG reporting tools by integrating superior embedding models for better understanding of ESG documents and communications.
Authors
Motaz Saad, Veronica Cretì, Ivan Gentile, Kianna Kazemi, Antonella Longo
Abstract
The use of Environmental, Social, and Governance (ESG) data is fundamental for modern corporate accountability, sustainability reporting, and financial decision-making. Embedding models have emerged as a powerful approach for transforming unstructured ESG text into numerical representations suitable for downstream natural language processing (NLP) tasks. However, their effectiveness in these ESG-specific tasks has not been systematically studied. In this paper, we construct a benchmark dataset specifically tailored to the ESG domain. We benchmark fourteen models, both open-source and closed-source embedding models, comparing their performance with respect to retrieval, and Retrieval-Augmented Generation (RAG). The results demonstrate performance variations across different models, with Qwen3-based models achieving the highest overall performance. This study provides practical insights into which models are better suited for ESG RAG tasks.