# Spatiotemporal Analysis of Electronic Cigarette Perception on Twitter/X Using Natural Language Processing

## TL;DR

This study analyzes how people's views on e-cigarettes changed over time and across countries on Twitter, using AI to process millions of tweets.

## Contribution

A novel spatiotemporal analysis of e-cigarette perception on social media using NLP and RoBERTa fine-tuning.

## Key findings

- E-cigarette-related tweets increased notably in the UK and Australia during the study period.
- Positive sentiment toward e-cigarettes was more common than negative, especially among users.
- Negative tweets highlighted health risks and youth harm, while positive tweets framed vaping as a smoking cessation aid.

## Abstract

Electronic cigarettes (e-cigarettes) have become popular in recent years, particularly among the youth and young adults. This study aims to examine the spatiotemporal patterns of online perception of e-cigarettes on Twitter/X.

Through the Twitter API (Application Programming Interface), over 3 million e-cigarette-related tweets were collected from March 11, 2021, to March 14, 2023, using related keywords, such as “e-cigarette” and “vaping”. After data cleaning (such as removing duplicates and retweets) and filtering, 2,140,439 non-commercial tweets were identified. Two human coders independently hand-coded 300 randomly selected tweets regarding relevance (yes or no), sentiment (positive, negative, or neutral), and whether the Twitter user is a likely e-cigarette user (yes or no). An additional 2,000 randomly selected tweets were single-coded. The labeled 2,300 tweets were used to fine-tune a pre-trained RoBERTa (Robustly Optimized BERT) model, which achieved good performance (F1 scores > 0.7). The Latent Dirichlet Allocation (LDA) method was used to identify the major topics in tweets with either positive or negative sentiment.

We observed a noticeable increase in the number of e-cigarette-related tweets, especially in the UK and Australia, during the study period. Nearly half of the tweets (49.7%, 1,063,317/2,140,439) were neutral. The proportion of tweets with a positive sentiment toward e-cigarettes was higher than that with a negative sentiment, at 27.0% vs. 23.3%. Except for Australia, in the US and UK, especially Canada, there were more positive tweets than negative ones. There was a rising trend in the proportion of tweets with a negative sentiment in the UK and Australia. Additionally, e-cigarette Twitter users were more likely to hold a positive sentiment toward e-cigarettes than non-users, 41.19% vs. 9.74%. Positive topics framed vaping as a desirable, emotionally driven alternative that supports smoking cessation, whereas negative topics emphasized health risks, youth harm, environmental concerns, and calls to quit despite perceived reduced harm.

Online perceptions of e-cigarettes on Twitter varied over time and across different countries. E-cigarette users and non-users held different sentiments toward e-cigarettes. Findings from this study provide timely monitoring in online perception of e-cigarettes on social media, offering valuable guidance for future tobacco regulations.