Bhatnagar, Shivansh (2025) Unveiling Regional Sentiment Shifts: Deep Learning and Explainable NLP in Reddit Mental Health Posts (UK and Ireland). Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (5MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (1MB) | Preview |
Abstract
This paper discusses the variation in emotional distress manifestation on Reddit in the United Kingdom and Ireland between prior to and post-covid-19. AI, NLP, and statistical learning were used to collect, clean, and analyze posts that were performed with a mixed manual/automated sentiment labeling process, with feature engineering. The problem of imbalances in sentiment and location was overcome by using SMOTE to provide balanced results. The methods of explainable AI (SHAP, LIME) helped the transparency whereas LDA found significant themes. The most accurate classification model scored 86% F1. Using auto ARIMA forecasting, an 18% sustained growth in distress-related messages in the post-pandemic era was estimated. providing real-world advice to mental health professionals and researchers on how to use advanced NLP and explainable AI to get real-time measurements of the overall sentiment and analyze the emotions and feelings associated with a given topic in the population.
Actions (login required)
![]() |
View Item |
Tools
Tools