A Transformer Based Model to Detect Environmental Related Distress on Online Social Media

Abstract

Online social media has been recently employed as the source of data for the detection of traffic-related activities and to control the flow of traffic in smart cities all over the world. Till now an ample amount of research work has been done in identification of traffic-related incidents with the help of online social media platforms (mostly Twitter). Along with the problem of traffic the society face the problem of increase in pollution and its related health hazards. Online social media proved to be effective in correctly mapping the sentiments of the people suffering from the problem of pollution with the exact measure of the harmful substances in the air. In this work, we presented a novel approach for analysing the environment-related distress on online social media related to pollution using BERT based transformer. In this paper we analysed the tweets and classified them into two classes: i) Pollution specific, ii) Location Specific by using the machine learning algorithms. From the location-specific tweets about the pollution, we then identify the location with serious pollution-related issues. The model was trained and evaluated on a dataset of 21,549 geotagged tweets collected from Delhi over a one-year period (January 2019–January 2020). The proposed BERT-based classifier achieved an accuracy of 96.5% and an F1-score of 96%, outperforming baseline machine learning algorithms and existing BERT-based benchmarks. Unlike prior BERT-based sentiment models that focus on general positive/negative polarity, our approach is specifically fine-tuned to detect pollution-related environmental distress and integrates geographic information extraction to identify urban pollution hotspots. This dual classification into Pollution Specific (PS) and Location Specific (LS) categories enables direct, actionable outputs for urban planners and traffic management authorities. The findings carry direct implications for smart city governance, enabling authorities to identify pollution hotspots and design data-driven interventions including traffic rerouting and targeted public health alerts.