Using Computer Vision to Detect E-cigarette Content in TikTok Videos.

Area of Science:

• Computational methods in public health
• Digital media analysis
• Machine learning applications in health surveillance

Background:

• Traditional text-based methods struggle with visual social media content.
• Platforms like TikTok are popular among youth and primarily visual.
• Detecting e-cigarette content on visual platforms requires advanced techniques.

Purpose of the Study:

• To develop and evaluate a computer vision model for detecting e-cigarette content on TikTok.
• To identify vaping devices, hands, and vapor clouds in TikTok images.
• To assess the accuracy and efficiency of automated detection methods.

Main Methods:

• Collected 826 images from 254 TikTok posts using 13 vaping-related hashtags.
• Annotated images for vaping devices, hands, and vapor clouds.
• Developed and trained a YOLOv7 computer vision model on 85% of images and tested on 15%.

Main Results:

• The computer vision model achieved a recall of 0.77 for vape devices, hands, and vapor.
• Vape devices were correctly classified with 92.9% accuracy.
• The model demonstrated an average F1 score of 0.81 for object detection.

Conclusions:

• Computer vision effectively detects e-cigarette content on visual social media platforms like TikTok.
• Automated detection of vaping devices and vapor clouds is feasible and accurate.
• These methods can support research and regulatory efforts for e-cigarette content online.