Sentiment Analysis of Weather-Related Tweets from Cities within Hot Climates

Abstract

Evidence exists that exposure to weather hazards, particularly in cities subject to heat island and climate change impacts, strongly affects individuals’ physical and mental health. Personal exposure to and sentiments about warm conditions can currently be expressed on social media, and recent research noted that the geotagged, time-stamped, and accessible social media databases can potentially be indicative of the public mood and health for a region. This study attempts to understand the relationships between weather and social media sentiments via Twitter and weather data from 2012 to 2019 for two cities in hot climates: Singapore and Phoenix, Arizona. We first detected weather-related tweets, and subsequently extracted keywords describing weather sensations. Furthermore, we analyzed frequencies of most used words describing weather sensations and created graphs of commonly occurring bigrams to understand connections between them. We further explored the annual trends between keywords describing heat and heat-related thermal discomfort and temperature profiles for two cities. Results showed significant relationships between frequency of heat-related tweets and temperature. For Twitter users exposed to no strong temperature seasonality, we noticed an overall negative cluster around hot sensations. Seasonal variability was more apparent in Phoenix, with more positive weather-related sentiments during the cooler months. This demonstrates the viability of Twitter data as a rapid indicator for periods of higher heat experienced by public and greater negative sentiment toward the weather, and its potential for effective tracking of real-time urban heat stress. Social media such as Twitter allow individuals to broadcast their opinions in real time, including perceptions and sensations related to weather events. Evidence from two cities exposed to hot weather—one equatorial and one desert subtropical—indicates that tweets were sensitive to seasonal temperature differences even within a small range. For Twitter users exposed to no strong temperature seasonality, generally negative sentiments to hot weather were seen year-round. In Phoenix with more pronounced seasonality, tweets were more positive in sentiment during the cooler months. This result shows promise for the medium as a rapid real-time indicator—or a snapshot—for societal sentiment to weather events.