An Application of Machine Learning and Textual Analytics for Visualisation and Enhanced Interpretation of Sustainability Disclosure

ABSTRACT Social and environmental accounting provides accountability mechanisms through which organisations communicate their sustainability practices to stakeholders. However, the increasing volume, complexity and diversity of sustainability disclosures make manual content analysis both labour‐intensive and susceptible to subjectivity. Machine learning and textual analytics offer promising approaches for systematically processing and visualising large qualitative datasets with greater consistency and transparency. This study explores how machine learning‐based textual analytics can be used to collect, analyse and visualise sustainability disclosure data in ways that enhance public accountability. More specifically, it focuses on how local governments use language to construct and communicate their sustainability commitments and performance. The empirical material consists of 237 annual reports from 79 local government areas in Victoria, Australia, covering the period 2015–2017. Drawing on a realist constructionist ontology, machine learning theory and stakeholder theory, the authors develop a machine learning‐based approach that transforms unstructured qualitative data into measurable insights. The methodology integrates VADER sentiment analysis, Latent Dirichlet Allocation topic modelling and Doc2Vec embeddings with cosine similarity to assess the comprehensiveness of sustainability disclosures. The analysis reveals that sustainability disclosures exhibit predominantly positive to neutral sentiment, with the dominant themes relating to financial and environmental challenges. The reports were categorised into three comprehensiveness levels, with the majority demonstrating a medium level of sustainability disclosure. Temporal analysis further suggests distinct patterns in how local governments frame sustainability initiatives and indicates that sustainability is conceptualised as a multidimensional construct. The study makes a methodological contribution by demonstrating how machine learning can enhance the rigour, scalability and transparency of sustainability accounting research. It also extends realist constructionist ontology into the domain of computational analysis and provides practical benchmarking tools for local governments. For stakeholders, the proposed visualisation techniques offer methods for assessing public sector sustainability accountability.