Identifying Topics for E-Cigarette User-Generated Contents: A Case Study From Multiple Social Media Platforms

Yongcheng Zhan, Ruoran Liu, Qiudan Li, Scott J. Leischow, Daniel ZengJournal of Medical Internet Research 2017 · DOI

TL;DR

This paper uses Latent Dirichlet Allocation (LDA) to analyze user-generated content about e-cigarettes across Reddit, JuiceDB, and Twitter, uncovering key topics like promotions, flavor discussions, experience sharing, and regulation debates. The findings highlight platform-specific discussion patterns and propose a unified model for leveraging social media data for e-cigarette policy development.

What Problem Does This Paper Address?

The rapid growth of the e-cigarette market has led to widespread discussions and debates, making it crucial for researchers and policymakers to understand public sentiment and emerging trends. Traditional methods like surveys are time-consuming and expensive. Social media platforms offer a rich source of user-generated content, but analyzing this vast amount of unstructured data presents a significant challenge. This paper aims to bridge this gap by demonstrating how to systematically mine and understand e-cigarette-related discussions across multiple social media platforms, providing actionable insights for policy and research.

Methods

The core methodology employed is Latent Dirichlet Allocation (LDA), a generative statistical model for unsupervised topic modeling. LDA is used to automatically discover hidden thematic structures within large collections of text documents. The authors applied basic Natural Language Processing (NLP) techniques, including tokenization, stop word removal, and stemming, to preprocess the collected data before feeding it into the LDA algorithm. They utilized the Python package gensim for LDA analysis and employed hierarchical Dirichlet process (HDP-LDA) to help determine the optimal number of topics. The process involved identifying key words associated with each topic and then categorizing these topics into broader themes.

Datasets

The study utilized data from three distinct social media platforms:

  • Reddit: 27,638 e-cigarette flavor-related posts from January 1, 2011, to June 30, 2015.
  • JuiceDB: 14,433 e-juice reviews from June 26, 2013, to November 12, 2015.
  • Twitter: 13,356 "e-cig ban"-related tweets from January 1, 2010, to June 30, 2015.

The authors note that while they focused on specific keywords for data collection, future work could expand this to include terms like "vape," "vapor," and "vaping."

Key Insights and Ideas

  • Platform Specialization: Different social media platforms serve distinct purposes and foster unique discussion patterns. Reddit acts as a comprehensive forum for a wide range of topics, while JuiceDB is highly specialized for e-liquid reviews, and Twitter excels at rapid information transmission and campaign coordination.
  • The Rise of E-Cigarette Promotions: A significant portion of user-generated content revolves around promotions, including direct sales from vendors and peer-to-peer trading among users. The prevalence of trading on Reddit (10.72% of posts) highlights a growing secondhand market that poses challenges for regulation and surveillance.
  • Flavor is King (and Complex): Flavor discussions are a dominant theme, with consistent categorization of flavors (e.g., fruit, cream) across platforms. This suggests a stable consumer preference landscape that could be monitored for emerging trends.
  • Regulation Debates are Fierce: E-cigarette regulation is a highly contentious topic, particularly on Reddit and Twitter. The analysis reveals a strong opposition to bans, often framed around personal freedom and the potential of e-cigarettes for smoking cessation.
  • A Unified Model for Policy Feedback: The paper proposes a "unified feedback model" that integrates topic modeling with social media data to provide policymakers with a more comprehensive and efficient way to gather public sentiment and feedback on policy proposals.

Key Results

  • Four Major Topic Categories: The LDA analysis identified four overarching topic categories across the platforms: promotions, flavor discussions, experience sharing, and regulation debates.
  • Cross-Platform Linkages: The study found promotion links between social media platforms, with a notable connection between JuiceDB and Reddit (87.30% of JuiceDB links pointing to Reddit).
  • Flavor Consistency: JuiceDB and Reddit identified consistent flavor categories, indicating a shared understanding of popular e-liquid flavors.
  • Vaping Methods Discussed: E-cigarette vaping methods and features like steeping, throat hit, and vapor production were broadly discussed on both Reddit and JuiceDB.
  • Reddit vs. Twitter for Regulation: Reddit was a primary platform for policy discussions, with a majority of posts (60.7%) expressing negative attitudes toward regulations. Twitter was more focused on launching campaigns using specific hashtags.
  • Significant Topic Distribution Differences: A chi-square test revealed a statistically significant difference (P<.001) in topic distribution between Reddit and JuiceDB, underscoring their distinct user perspectives.
  • Twitter Campaign Themes: Analysis of Twitter hashtags revealed themes against e-cigarette regulations, including arguments about "no harm," "smoking cessation," "personal freedom," and "pharma interests."

Limitations and Future Work

Limitations:

  • Limited Platform Scope: The study focused on Reddit, JuiceDB, and Twitter. Other platforms like Facebook and dedicated e-cigarette forums could provide additional insights.
  • Keyword Specificity: Data collection was based on specific keywords. A broader keyword set, including terms like "vape," "vapor," and "vaping," could yield more comprehensive results.
  • Lack of Demographic Data: The absence of reliable demographic information (age, gender) prevented subgroup analysis to understand differing patterns among user groups.
  • Single Analysis Method: Only LDA was used for topic identification. Other data mining tools, such as sentiment analysis, could offer deeper insights, especially for regulation-related posts.

Future Work:

  • Extend Platform Coverage: Incorporate data from other social media platforms.
  • Refine Keyword Strategy: Use a more comprehensive set of keywords for data collection.
  • Incorporate Demographic Data: If possible, find ways to infer or collect demographic information for more nuanced analysis.
  • Employ Additional Data Mining Techniques: Conduct sentiment analysis and explore other machine learning approaches.
  • Modify and Extend LDA: Explore advanced LDA techniques, such as supervised learning or graph-based methods, for more precise analysis.
  • Investigate Emerging Phenomena: Further explore topics like individual trading, vendor promotions, and automatic e-liquid detection.
  • Analyze Platform Interconnections: Study how information flows and is shared across different platforms.
  • Address Bots and Spam: Develop methods to filter out non-authentic content on platforms like Twitter.

Summary updated

Abstract

This study examined Reddit, JuiceDB, and Twitter as social media data sources for e-cigarette research. These mined findings could be further used by other researchers and policy makers. By utilizing the automatic topic-modeling method, the proposed unified feedback model could be a useful tool for policy makers to comprehensively consider how to collect valuable feedback from social media.
BibTeX
@article{zhan2017identifying,
  author = {Yongcheng Zhan and Ruoran Liu and Qiudan Li and Scott J. Leischow and Daniel Zeng},
  title = {{Identifying Topics for E-Cigarette User-Generated Contents: A Case Study From Multiple Social Media Platforms}},
  journal = {Journal of Medical Internet Research},
  year = {2017},
  doi = {10.2196/jmir.5780},
}
Figure