From Dictionaries to Deep Learning: A Systematic Mapping Review of the Natural Language Processing Tasks Used to Analyze Sustainability Reports

ABSTRACT This study aims at shedding light on the vast landscape of natural language processing (NLP) tasks used when analyzing sustainability reports or sustainability within integrated annual reports. A systematic literature review is carried out, identifying 160 studies of relevance. The results are presented as a hierarchical categorization, grouping methods and models into tasks, and tasks into categories. Most common tasks are keyword searches, (computer aided) content analysis, and topic modeling. Researchers apply NLP most frequently in a comparative setting, investigating relations or exploring changes over time. Common limitations are different linguistic and cultural features, temporal changes in reporting language, and the risk to introduce or amplify bias. This study provides researchers and practitioners with guidance and recommendations when applying NLP tasks. It additionally proposes a research agenda focused on guidelines and benchmarks as well as replication and verification studies.