Küçükali Hüseyin, Erdoğan Mehmet Sarper
Centre for Public Health, Queen's University Belfast, Belfast, UK.
Research Center for Healthcare Systems and Policies, Istanbul Medipol University, Istanbul, Türkiye.
Nicotine Tob Res. 2025 May 22;27(6):988-996. doi: 10.1093/ntr/ntae276.
Tobacco companies use social media to bypass marketing restrictions. Studies show that exposure to tobacco promotion on social media influences subsequent smoking behavior, yet it is challenging to monitor such content. We developed an artificial intelligence that can automatically identify tobacco-promoting content on social media.
In this mixed methods study, 177,684 tobacco-related tweets published on Twitter in Turkish were collected. Through inductive content analysis of a sample of 200 tweets, the main mechanisms by which tobacco is promoted on social media were identified. Then, a sample of 5000 tweets was deductively analyzed and labeled based on those mechanisms. A pre-trained transformer-based Large Language Model was fine-tuned using the labeled dataset. Then, tobacco promotion in all tweets was predicted using this model.
The main mechanisms of tobacco promotion on social media included modeling the behavior, expressing positive attitudes, recommending use, and marketing brands or vendors. The developed model identified tobacco-promoting social media content with 87.8% recall and 81.1% precision. The utility of the model was demonstrated in the analysis of tobacco promotion in tweets for a period of a month.
This tool makes it possible to monitor tobacco promotion in social media and creates new opportunities for tobacco control policy and practice, not only in surveillance and enforcement but also in health promotion.
Tobacco promotion in social media is a well-known yet hard-to-addressed problem due to the nature of social media. This study leverages a cutting-edge AI approach, Large Language Models, to identify tobacco promotion in social media content automatically and precisely. The developed model offers better prediction performance than previously proposed techniques. The study enables surveillance of tobacco-promoting content both for research purposes and enforcement of tobacco control measures. Furthermore, we suggest a range of health promotion opportunities this tool can help with from developing personal skills to creating supportive environments and strengthening community actions.
烟草公司利用社交媒体绕过营销限制。研究表明，接触社交媒体上的烟草促销活动会影响随后的吸烟行为，但监测此类内容具有挑战性。我们开发了一种人工智能，它可以自动识别社交媒体上的烟草促销内容。
在这项混合方法研究中，收集了在推特上发布的177684条土耳其语烟草相关推文。通过对200条推文样本进行归纳性内容分析，确定了社交媒体上推广烟草的主要机制。然后，基于这些机制对5000条推文样本进行演绎分析并标注。使用标注数据集对预训练的基于Transformer的大语言模型进行微调。然后，使用该模型预测所有推文中的烟草促销情况。
社交媒体上推广烟草的主要机制包括行为示范、表达积极态度、推荐使用以及营销品牌或供应商。所开发的模型识别烟草促销社交媒体内容的召回率为87.8%，精确率为81.1%。该模型的效用在对一个月期间推文中的烟草促销分析中得到了证明。
该工具使监测社交媒体上的烟草促销成为可能，并为烟草控制政策和实践创造了新机会，不仅在监测和执法方面，而且在健康促进方面。
由于社交媒体的性质，社交媒体上的烟草促销是一个众所周知但难以解决的问题。本研究利用前沿的人工智能方法，即大语言模型，自动、精确地识别社交媒体内容中的烟草促销。所开发的模型比先前提出的技术具有更好的预测性能。该研究使我们能够出于研究目的和执行烟草控制措施对烟草促销内容进行监测。此外，我们提出了一系列该工具可助力的健康促进机会，从培养个人技能到营造支持性环境以及加强社区行动。
