Non-Brazilian environmental perception on the Amazon rainforest: an approach using text mining from social media

Background: The Amazon rainforest plays an essential role in sheltering global biodiversity and providing essential ecosystem services. However, the region has been threatened by increased rates of deforestation and degradation, which is often reported as the failure of the Brazilian conservation policy that could affect the image of agriculture Brazilian products abroad. In this sense, here we aimed to evaluate the foreign environmental perception of the region using data from social medias, to assess which are the main terms related to the region, how their importance varies over time and whether this perception tends to be more negative or positive. For this, we used data from posts made on Twitter (R) involving the term "Amazon rainforest" over 18 weeks using a data mining process. From these posts, we extracted the text, which went through steps of cleaning and organization, as well as crossing with additional databases (such as the sentiment dictionary). From the final data set, we first evaluated which are the main terms present in these posts and how their importance varied in the evaluated period. Next, we assess whether the terms cited are mainly negative or positive and how this sentiment varied over the monitoring period.Results: We found that the main terms are related to the environmental context such as "carbon", "deforestation" and "fire" and that there is a relationship between their importance and the occurrence of fires in the region, which is positive with a temporal lag. In addition, we also found that the posts are, on average, composed of 68.47% of negative terms and that this sentiment predominates throughout the entire time series, being higher during the fire peaks in the region.Conclusion: Our results indicate that the environmental perception of the region presented in the posts in Twitter (R) is mainly negative, due to the degradation and fires observed over the last few years. We also discuss limitations of the approach and establish perspectives for future work.