WildFireCan-MMD: A Multimodal Dataset for Classification of User-Generated Content During Wildfires in Canada

The introduction of the WildFireCan-MMD dataset marks a significant advancement in the classification of user-generated content during wildfires in Canada. With traditional data sources proving slow and costly, this dataset, consisting of X posts, offers a timely solution for accessing critical information during emergencies. The research highlights the effectiveness of custom-trained models, which achieved an impressive 84.48% f-score, surpassing both baseline classifiers and zero-shot vision-language models. This finding emphasizes the necessity of tailored datasets and task-specific training in enhancing disaster response capabilities. As wildfires become increasingly prevalent, the ability to analyze and extract insights from social media data is vital for improving response strategies and understanding trends, ultimately aiding in better management of wildfire situations.