Using social media and machine learning to understand sentiments towards Brazilian National Parks

Abstract

Protected areas (PAs) play a vital role in the conservation of natural and cultural heritage while supporting local livelihoods. However, in Brazil, where limited resources and poor effectiveness lead to negative sentiments and are leveraged as criticism towards PAs, it is necessary to better comprehend public perceptions of Brazilian PAs and identify the key factors contributing to negative sentiments. Here, we use data from online discussions about Brazilian national parks (NPs) on Twitter and sentiment analysis to explore this question. We classified the sentiment of ∼100,000 tweets collected over a twelve-year period (2011−2022) using the BERTimbau Base model. We also performed a topic modelling with the BERTopic model to identify prevalent subjects concerning Brazilian NPs. We identified 18,388 (17.30 %) posts expressing negative sentiment towards NPs, mostly associated with wildfires occurring between 2011 and 2017 and concerning government decisions impacting conservation efforts after 2019. The results revealed six prominent topics: (1) Wildfires; (2) Security; (3) Regulations; (4) Wildlife roadkill; (5) Privatization; (6) Lack of financial resources, reflecting a diverse range of negative sentiments regarding the parks, surpassing isolated events. Furthermore, examining specific topics on a per-park basis proved beneficial in identifying distinct issues and conflicts in the five most tweeted NPs, facilitating targeted conservation actions. Using social media data to better understand public perceptions of NPs can strengthen their management and governance by reinforcing their conservation initiatives and enhancing visitor experiences. Our findings underscore the value of sentiment analysis in identifying gaps and driving improvements in the management of protected areas.