Towards Sentiment Analysis of Tobacco Products’ Usage in Social Media

Proceedings of Recent Advances in Natural Language Processing, pages 1545–1552 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_173 1545 Towards Sentiment Analysis of Tobacco Products’ Usage in Social Media Venkata Himakar Yanamandra ∗, Kartikey Pant ∗and Radhika Mamidi International Institute of Information Technology, Hyderabad, India {himakar.y, kartikey.pant}@research.iiit.ac.in , radhika.mamidi@iiit.ac.in. Abstract Contemporary tobacco-related studies are mostly concerned with a single social media platform while missing out on a broader au- dience. Moreover, they are heavily reliant on labeled datasets, which are expensive to make. In this work, we explore sentiment and product identification on tobacco-related text from two social media platforms. We release SentiSmoke-Twitter and SentiSmoke- Reddit datasets, along with a comprehensive annotation schema for identifying tobacco products’ sentiment. We then perform bench- marking text classification experiments us- ing state-of-the-art models, including BERT, RoBERTa, and DistilBERT. Our experiments show F1 scores as high as 0.72 for sentiment identification in the Twitter dataset, 0.46 for sentiment identification, and 0.57 for product identification using semi-supervised learning for Reddit. 1 Introduction Smoking tobacco causes more than 8 million deaths each year1. While cigarette smoking has fallen globally, e-cigarettes remain the commonly used tobacco product among the US youth2. So far, 2, 807 cases and 68 deaths have been recorded due to e-cigarette use-associated lung injury in the US2 alone. These developments prompted the US government to enforce a coun- trywide ban on addictive flavors and sales to mi- nors. As smokers are more likely to develop severe COVID-19 symptoms, WHO has introduced the world’s first digital health worker Florence based on AI to fill the gap between overburdened medical system users who are trying to quit3. Health-related work.∗ The first two authors have contributed equally to the 1http://bit.ly/WHO_tob 2https://www.cdc.gov/tobacco/ 3https://bit.ly/who_AI discourses on social media have risen substantially over the years (Tamersoy et al., 2015) since social media provides an opportunity to informally ex- press opinions freely with like-minded individuals across geographical and societal barriers. Twitter boasts of 330 million active monthly users who send out half a billion tweets daily4 while Reddit has 430 million monthly users and has 100k+ communities 5. Reddit provides ex- pendable and pseudo-anonymous accounts that are well suited for controversial discussions, including the perception of electronic cigarettes and mari- juana, which might be inappropriate to discuss on non-anonymous forums (Park and Conway, 2018). In Salathé and Khandelwal (2011), we see that information flows more often between users who share the same sentiments. Studying trends of first- person accounts of experience and sentiment to- wards different tobacco products in these forums becomes imperative in the ongoing bio-surveillance and regulatory efforts (Kim et al., 2015; Pant et al., 2019). Real-time monitoring of public sentiment and informativeness gives us an opportunity for bottom-up discovery of emergent patterns, espe- cially in vulnerable and ethnically diverse popu- lations that may not be readily detectable by tra- ditional methods (Myslín et al., 2013; Lienemann et al., 2017). Although Yanamandra et al. (2020) identified tobacco products on Twitter, they leave a gap in identifying the products’ sentiment. This work explores …

Content truncated to 500 words