ESG rating employing NLP and multi-criteria ranking techniques for Russian corporations

L. A. Myl'nikov ; M. A. Storchevoy ; V. V. Lapina ; A. A. Murach

The Bulletin of Irkutsk State University. Series Mathematics, Tome 50 (2024), pp. 125-142

Résumé

The significance of the research outlined here lies in the challenge of selecting partner companies under tight timeframes and with limited access to comprehensive data, particularly concerning a wide range of ethics and sustainability issues. This study aims to address this challenge by developing a model for the automated processing and analysis of textual information concerning groups of companies, particularly within the context of the Environmental, Social, and Governance (ESG) concept. The study incorporates methods for automatically extracting topics from textual data, utilizing machine learning techniques, conducting multi-criteria ranking, and employing both comparative and expert analyses of the results. To conduct our experiments, we compiled a dataset comprising over 1200 reports from leading Russian companies spanning the years 2019 to 2022. Additionally, we leveraged news articles posted on the FORBES-Russia website. As a result of our work, we have developed a model and methodology for analyzing textual information about groups of companies, facilitating their ranking. The analysis of the obtained results, both qualitatively and quantitatively, demonstrates their non-random and justified nature.The study showcases the efficacy of the proposed model in selecting companies through the ranking of a limited set of candidates based on accessible textual information. The insights from this research are valuable not only for rating agencies but also for companies seeking to conduct their own analysis when choosing partners.

Détail

Keywords: ESG, rating, ranking, natural language processing (NLP), model, company reports, Multi-Attribute Utility Theory (MAUT), topic extraction, word dictionary, feature matrix.