Demographic trends in e-cigarette social media marketing: perceiving gender presentation and facial age via computer vision.

Introduction: Demographic characteristics of individuals featured in tobacco promotions, such as gender presentation and perceived age, can influence the impact of tobacco marketing on young people. These characteristics often must be estimated when analyzing social media posts. A novel approach for tobacco control research is to harness computer vision models to identify faces in images or videos and then estimate gender and age presentation based on facial features. Such models could facilitate monitoring trends in the demographics of e-cigarette promoting-content when self-report data are unavailable.

Methods: We trained computer vision models to identify faces using the WiderFace dataset (32,203 images), and to estimate gender and age presentation given a detected face using the UTKFace dataset (10,670 images). We then applied our models to a collection of 69,788 Instagram posts from 230 e-cigarette influencers over 2019-2022 to assess temporal trends in demographics.

Results: The best performing models were DINO for face detection, ConvNext-v2 for gender presentation (96.7% accuracy), and Eva02 for age estimation (70% accuracy). Analyzing 58,535 detected faces across 98.3% of influencer accounts, we observed a significant shift in the gender distribution of e-cigarette promoting posts on Instagram, with 50% female at the study start (2019) falling to 31% female by the study end (2022). The majority of posts (68%) showed individuals in the 12-24 age range, a stable trend.

Conclusions: Computer vision models measured gender and age presentation through facial analysis, enabling scalable demographic trend monitoring of e-cigarette marketing on social media.

Implications: Analyzing 69,788 Instagram e-cigarette influencer posts for gender and age presentation using facial recognition algorithms, we detected 58,535 faces and found a trend from equal gender representation in 2019 posts shifting down to 31% female prevalence by 2022. The majority (68%) of posts featured adolescents and young adults of age 12 - 24 and this trend was stable. These findings reinforce the need for expanded theory development of moderation and mediation effects of gendered and age-related tobacco marketing strategies, while highlighting the power of computer vision to scalably monitor real-world tobacco communication and inform regulatory policy.