A Text-Based Approach to Sustainability Indicators

Abstract

This dissertation explores natural language processing (NLP) techniques applied to the field of sustainability, with a focus on the Sustainable Development Goals (SDGs) and corporate sustainability reporting. It is divided into four main sections, consisting of an introduction, two main research sections and a conclusion. The research takes advantage of advances in large language models, in particular those developed from BERT (Bidirectional Encoder Representations from Transformers) and its subsequent variants, to develop methods for analysing and extracting information from diverse textual sources. The overall aim is to develop sustainability indicators that can provide a good alternative to existing measures. The first main topic of this thesis focuses on sustainable development, introducing new approaches to quantify research and information on the SDGs, and creating a research attention index based on academic articles and a sentiment index based on voluntary national reviews. Both indices will be compared with the official SDG scores provided by the United Nations. The second main topic focuses on sustainability reporting, uncovering possible selective disclosure and creating an ESG sentiment index to compare with ESG ratings provided by rating agencies. The thesis provides a comprehensive view of how NLP can be used to develop indicators and tools that support the efficient extraction and analysis of sustainability-related data, providing valuable resources for ongoing research and policy-making in this area.