Mapping Emotional Expressions in Social Media Discourse : A Spatial Analysis of Public Perspectives on Urban Water Systems

Sammanfattning

Urban water systems face growing challenges from climate change, ageing infrastructure, and public concern about water-related risks. Social media offers insight into how citizens emotionally respond to these challenges and how such responses may inform policy and planning. This thesis analyses online discourse about urban water systems in Stockholm and Bursa using five emotion categories: anger, fear, mistrust, hope, and neutral. The study applies a mixed-methods framework to examine online water discourse across emotional, thematic, spatial, and contextual dimensions. A dataset of 1,976 comments from Twitter/X, Reddit, YouTube, Flashback, and Ekşi Sözlük was collected for 20222025. To identify the most suitable approach for classifying emotions in multilingual social media comments, three methods were evaluated: supervised machine learning, zero-shot XLM-RoBERTa, and a hybrid lexicon-based model. Linear SVM achieved the strongest performance and was applied to the full corpus. BERTopic identified themes, and the results were visualised through a Leaflet-based map as a prototype socialemotional layer for exploratory urban water governance analysis. The findings show that the two cities differed not simply in overall emotion levels, but in the dominant framing of water-related concerns. Stockholm discourse was mainly characterised by mistrust, institutional criticism, and technical reasoning, while Bursa discourse was dominated by fear linked to water scarcity, crisis experience, and environmental conditions. Spatial analysis showed broader fear-oriented patterns in Bursa and more heterogeneous location-specific patterns in Stockholm. Overall, the thesis shows how multilingual NLP and spatial visualisation can be used to identify emotional, thematic, and place-related patterns in online discourse about urban water systems. The interactive map integrates these results as a prototype socialemotional layer for exploratory analysis of public concerns.