Sentiment Analysis in Social Medias for Threats Prediction of Natural Extreme Events

Abstract

This chapter presents a multidisciplinary solution that considers as evolution of endogenous natural extreme event deforestation the threats of droughts and fires in the Brazilian Amazon region. The data are collected from social media, such as newspapers and magazines, related to the domain of droughts and fires that could trigger and accelerate the process of deforestation in the period from 2015 to 2020. The data science concepts and natural language processing with sentiment analysis are used and generate the degree of threat that each news presents regarding the possibility of deforestation. This threat degree generates an endogenous time series that will be used to predict the threat evolution of occurrence of drought, fire, and deforestation for a future of three months. The time series prediction is performed using machine learning and deep learning with an LSTM neural network. An analysis of the endogenous time series is performed using the statistical tools of mean, variance, standard deviation, asymmetry, and kurtosis.