ANALYZING CITIBANK'S ESG REPORTS USING THE LATENT DIRICHLET ALLOCATION (LDA) MODEL

Abstract

This article applies Latent Dirichlet Allocation (LDA), an unsupervised machine learning technique, to analyze Citibank’s Environmental, Social, and
Governance (ESG) reports from 2019 to 2023. The goal is to discover underlying topic structures and understand how sustainability communication has changed
over time. Six research questions are developed to extract and compare the major themes in each yearly report, followed by a cross-year analysis. The findings
show a progressive shift in theme emphasis, from broad issues like corporate governance, employee involvement, and community support in previous years to
more particular concerns like climate-related financial risk, sustainable financing, and emission verification in recent years. These findings demonstrate Citibank's
growing emphasis on measurable sustainability performance and regulatory compliance. The study demonstrates the efficacy of topic modeling in longitudinal
assessments of corporate ESG disclosures, providing insights into strategic communication trends and altering organizational priorities.