The dataset is described in "Fire Detection for Emergency Responders using X" by Stefanopoulos et al. (DOI: 10.5281/zenodo.12065460).


english_data, greek_data: json files with the columns ['id','ground_truth','Image-based Fire Detection','Concept-based Detection','Relevance Estimation']

    'id' : the post id as given from the X API
	
    'ground_truth': our annotation  using the guidlines.txt
	
	'Image-based Fire Detection': Probability of the post's image having a fire scored by the Image-based Fire Detection module.
	
	'Concept-based Detection': Probabilities of the post's image having the 10 concepts(Explosion_fire, Truck, Fire_Truck, Outdoor, Fields, Forest, Helicopter_Hovering, Landscape, Trees, and Vegetation) scored by the Concept-based Detection module.
	
	'Relevance Estimation': Probability of the post's text to be relevant for emergency responders scored by the Relevance Estimation module.
	
folds_en, folds_gr: The 5 crossvalidation folds used across the paper for reference. 

Questions at distef@iti.gr.