Research Summary: Quantifying Organizational Environmental Action from Web Data and Large Language Models

Research details a new computational framework leveraging large language models and web crawling to quantify organizational environmental actions from publicly available web content. The method addresses the challenge of dispersed and unstructured data, demonstrating its utility by creating a national database of Jewish congregations and analyzing their environmental activities based on their websites.

Why it matters

This research introduces a novel methodology for objectively assessing organizational environmental commitments and activities through publicly available data. Such a capability is crucial for stakeholders to independently verify and benchmark sustainability efforts, thereby influencing investment decisions, regulatory oversight, and public perception.

Key insights

• A scalable computational framework has been developed to convert unstructured organizational web content into structured metrics of environmental action.
• The framework effectively handles information dispersed across multiple webpages and communicated via unstructured text.
• The method was successfully applied to U.S. Jewish congregations, creating a national database of 4,964 congregations, with 2,657 having active, crawled websites.
• The process yielded a corpus of 154,454 webpages for analysis, demonstrating the approach's capability to process large datasets.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.16627