Examining the perspectives of environmental inspectors and planners through a developed Natural Language Processing pipeline: A case study of domestic wastewater treatment systems in the Republic of Ireland

Understanding the perspectives of local authority planners and environmental inspectors responsible for inspecting and regulating sources of rural water contamination (e.g., domestic wastewater treatment systems or DWWTSs) is a crucial step towards advancing environmental risk management. These officials offer unique first-hand insights into practical challenges, policy limitations, and ground-level realities, however their qualitative observations are often neglected in systematic analyses. With the increasing availability of rich text-based data, there is a clear need for analytical frameworks that can simplify and transform unstructured informational content into actionable, policy-relevant knowledge for environmental risk mitigation. Accordingly, the current study sought to integrate traditional Natural Language Processing (NLP) techniques with machine learning methods to analyse qualitative textual data. A total of 2539 interview paragraphs extracted from 20 interviews with domestic wastewater treatment system (DWWTS i.e., septic tank system) inspectors and planners across 12 Irish local authorities were pre-processed and encoded. Non-Negative Matrix Factorization (NMF) was employed to extract topic-based features, and subsequently clustered using unsupervised techniques. K-Means clustering produced the most coherent and interpretable results (Silhouette Score = 0.6136), identifying three distinct themes: Challenges, Achievements, and Recommendations. Sentiment analysis was subsequently performed using a fine-tuned RoBERTa model to explore emotional tone within each theme. Challenges were predominantly negative (496, 39.1 %), often highlighting resource limitations, administrative burdens, and public resistance. Achievements were primarily neutral (60.9 %) or positive (25.5 %), reflecting improved inspection practices and stakeholder engagement. Recommendations were also neutral (76.8 %), with 81 (16.5 %) reflecting optimism or constructive suggestions. NLP results highlight LA capacity constraints, desire for policy improvements and system grants as dominant themes relating to DWWTS management among LA officials. These findings may carry a wider resonance for research in environmental management performance and policy in modern, predominantly rural LAs.