Citation

Kuntsche E, He Z, Bonela AA, Riordan B. Int. J. Drug Policy 2023; 118: e104098.

Copyright

(Copyright © 2023, Elsevier Publishing)

DOI

10.1016/j.drugpo.2023.104098

PMID

37352767

Abstract

With the advent of social media sites, streaming services, and mobile devices, people are spending more time accessing digital media than ever before. Unfortunately, the increase in digital media use also signifies an increased exposure to representations of alcohol, tobacco, and other substances. Indeed, the portrayal of and references to alcohol and other substances occur frequently on digital media platforms via content sharing, posting, streaming, and marketing. For example, an estimated 2% of Tweets reference alcohol (Alhabash et al., 2018), and tobacco and alcohol are depicted in the majority of popular Netflix/Amazon Prime series (Barker et al., 2019). Social media platforms also show psychoactive substances such as e-cigarettes, cannabis, opioids, and prescription drugs with content that encourages or even normalizes the use of these substances (Suarez-Lledo & Alvarez-Galvez, 2021). Given the clear link between exposure to alcohol and other substances in the media and their increased use (Sargent & Babor, 2020), the WHO-UNICEF-Lancet commission report highlighted the need to better understand and reduce the amount that people are exposed to those harmful products (Clark et al., 2020).

Thus, identifying and monitoring exposure to and marketing of alcohol and other substances in digital media is extremely important to help inform the public and policy makers to make better health decisions. However, monitoring the portrayal of substances is challenging due to (i) the sheer amount of ever-growing content in digital media, and (ii) the time and resource restrictions of current data extraction and pre-processing methods, including manual annotation of substance-related images and text in digital media. Monitoring substance portrayal in digital media has become more feasible with the use of natural language processing and computer vision through deep learning, a domain of artificial intelligence (AI) that teaches computers to process and understand information like the human brain. This technology can automatically process large amounts of data in similar ways to humans, albeit more efficiently. For example, in our previous work, we have demonstrated that deep learning algorithms can automatically identify alcoholic beverages from a variety of context-related 'real-life' images and can screen all kinds of digital media for images of alcohol and potentially other substances free from any response or coding burden and with a relatively high accuracy (Bonela et al., 2022; Kuntsche et al., 2020).

Language: en