Development of a Deep Learning Technique to Analyze Air Quality from Social Media Response

Abstract

One of the prominent new-age methods used by today’s population in spreading awareness or drawing attention on an issue or concern is through social media platforms. In this study, the responses of general public to air quality that they share on a popular social media platform – Twitter, was taken as a virtual quantity that helps in measuring and analyzing the prevalent air quality. Machine learning technique based on self attention network was used to sort, clean and classify large amount of air pollution related Twitter responses extracted during 2019-2020 at Delhi in India. The temporal correlation of tweet response volumes with the ground monitored concentrations of air pollution and word cloud analysis were used to analyze the attitude of the public towards urban air quality. These analyses lead to the development of an air quality prediction model using ‘spaCy’ – similarity analysis method which attempts to predict the ground concentration range of pollutant PM2.5 from tweets received on a particular day.