Table 2.

Performance metrics for video classification for each model. The residual network with 50 layers (ResNet50) model performed consistently well across all 5 themes compared with the other 2 models.

	F1-scores
	General vape (theme 1)	Vaping cessation (theme 2)	Vape product marketing (theme 3)	TikTok influencer (theme 4)	Vape brands (theme 5)
ResNet50	0.974	0.972	0.960	0.975	0.975
VGG16a	0.932	0.986	0.953	0.955	0.980
ViTb	0.943	0.941	0.847	0.986	0.778

^aVGG16: Visual Geometry Group model with 16 layers.

^bViT: vision transformer.