Automated Detection of Digital Alcohol Marketing Using SCANNER: An Integrated Deep-Learning Approach.

INTRODUCTION

Alcohol marketing significantly influences consumption patterns, particularly among youth, heavy drinkers and women. Digital platforms have amplified this impact through targeted, immersive campaigns. However, monitoring such marketing remains a challenge due to its opaque and dynamic nature. This study introduces SCANNER Alcohol, an AI-enabled system designed to detect alcohol marketing in online content at the brand level.

METHODS

SCANNER Alcohol integrates object detection (logos) and optical character recognition (text) to automatically identify 134 alcohol brands within image and video data. The system was trained using annotated datasets of brand logos and validated through standard machine learning metrics. Real-world performance was assessed using social media screen recordings (brand accounts and 12 h of general use), benchmarked against manual coding.

RESULTS

SCANNER Alcohol achieved high algorithmic accuracy, with a mean average precision of 0.94, recall of 0.96 and F1 score of 0.95. In real-world testing, the model demonstrated strong performance, correctly identifying 98.9% of alcohol-branded posts in social media videos. SCANNER Alcohol achieved a low 6.7% false discovery rate, indicating high precision and low noise in the detection output.

DISCUSSION AND

CONCLUSIONS

SCANNER Alcohol is the first system to combine brand-level object and text detection for automated digital alcohol marketing surveillance. Its high accuracy on real-world data and ethical design make it a valuable tool for public health monitoring. SCANNER Alcohol offers a scalable, adaptable and ethically sensitive solution to support regulatory efforts to hold alcohol companies accountable for their digital marketing practices and to ultimately reduce alcohol-related harm.