Neurosymbolic AI for Mining Public Opinions about Wildfires

Cuc Duong, Vethavikashini Chithrra Raghuram, Amos Lee, Rui Mao, Gianmarco Mengaldo, Erik Cambria

Abstract: Wildfires are among the most threatening hazards to life, property, well-being, and the environment. Studying public opinions about wildfires can help monitor the perception of the impacted communities. Nevertheless, wildfire research is relatively limited compared to other climate-related hazards. This article presents our data mining work on public opinions about wildfires in Australia from 2014 to 2021. Three key aspects are analyzed: the topic of concern, sentiment polarization, and perceived emotions. We propose a data filtering approach to acquire golden samples to train a supervised model for emotion quantification to achieve the last target. The results show that the new model produces a more accurate emotion estimation than the existing lexicon approach. Through data analysis, we find that people have seen wildfires as one of the impacts of climate change; trends of tweets can reflect the damage of wildfires in real life.