Using Social Media Data to Understand the Impact of Promotional Information on Laypeople’s Discussions: A Case Study of Lynch Syndrome

Jiang Bian, Yunpeng Zhao, Ramzi G. Salloum, Yi Guo, Mo Wang, Mattia Prosperi, Hansi Zhang, Xinsong Du, Laura J Ramirez-Diaz, Zhe He, Yuan SunJournal of Medical Internet Research 2017 · DOI

TL;DR

This study leverages social media data (Twitter) to analyze how promotional health information about Lynch syndrome influences public discussions and attitudes. By employing topic modeling and sentiment analysis, the research reveals that promotional content significantly impacts laypeople's conversations, particularly during awareness events, offering valuable insights for public health communication strategies.

What Problem Does This Paper Address?

The increasing use of social media by various stakeholders (pharmaceutical companies, government agencies, health organizations) for health promotion raises a critical question: what is the actual impact of this information on the public's understanding and discussion of health issues? Specifically, for diseases like Lynch syndrome, a hereditary disorder increasing cancer risk, it's unclear how promotional campaigns translate into laypeople's awareness, attitudes, and conversations. This paper aims to bridge this gap by examining the relationship between promotional Lynch syndrome information on Twitter and laypeople's discussions, and how awareness events influence these interactions.

Methods

The study employed a multi-step approach:

  1. Data Collection and Preprocessing: Tweets related to Lynch syndrome were collected from Twitter using a set of keywords and a snowball sampling process. Non-English tweets were filtered out, and text was standardized (e.g., hashtags, URLs).
  2. Tweet Categorization: A two-step process was used to classify tweets into "unrelated," "promotional Lynch syndrome-related information," and "laypeople's discussions." This involved experimenting with both a Convolutional Neural Network (CNN) and a rule-based classifier, with the rule-based classifier showing better performance.
  3. Topic Modeling: Latent Dirichlet Allocation (LDA) was used to identify the main discussion themes within the Lynch syndrome-related tweets. The optimal number of topics was determined through a combination of statistical methods and qualitative analysis by coders.
  4. Sentiment Analysis: A CNN classifier was trained on a subset of laypeople's tweets to determine their sentiment (positive, negative, or neutral) towards Lynch syndrome.
  5. Analysis and RQ Answering: The results from topic modeling and sentiment analysis were used to answer three research questions concerning the most discussed topics, the influence of promotional information on discussions, and the impact of awareness events.

Datasets

The primary dataset consisted of 16,667 tweets collected from Twitter between October 26, 2016, and August 11, 2017. After preprocessing and filtering, 14,564 English tweets remained. A subset of 1092 tweets was used for sentiment analysis training, and 1000 tweets were used for developing the classification models. No new benchmarks were introduced; existing Twitter data was utilized.

Key Insights and Ideas

  • Social Media as a Health Communication Amplifier: The study strongly suggests that social media platforms like Twitter are powerful tools for disseminating health information and can significantly influence public discourse.
  • Promotional Content Drives Discussion: Promotional Lynch syndrome information on Twitter shows a strong correlation with laypeople's discussions, indicating that targeted campaigns can effectively engage the public.
  • Awareness Events Have a Measurable Impact: Specific events like Colon Cancer Awareness Month and Lynch Syndrome Awareness Day demonstrably increase tweet volume and can positively influence laypeople's attitudes, at least temporarily.
  • Topic Overlap is Key: The similarity in topic distributions between promotional content and laypeople's discussions highlights the direct influence of the former on the latter.

Key Results

  • Promotional vs. Laypeople's Discussions: Of the Lynch syndrome-related tweets, 81.43% were classified as promotional, and 18.57% as laypeople's discussions.
  • Top Topics: The most discussed topics were "treatment" (n=4080), "genetic testing" (n=3073), and "awareness/awareness events."
  • Sentiment: Overall, most laypeople's tweets were neutral (78.07%), with positive tweets (18.42%) significantly outnumbering negative ones (3.51%). However, the "treatment" topic had a higher proportion of negative tweets (16.67%).
  • Correlation: Laypeople's discussions showed a strong correlation with promotional information on the "awareness" topic (r=.989, P<.001) and moderate correlations on "screening," "genetic testing," "treatment," and "risk."
  • Awareness Event Impact: Tweet volume dramatically increased during March 2017 (Colon Cancer Awareness Month) and peaked around Lynch Syndrome Awareness Day (March 22, 2017). Sentiment analysis showed a temporary increase in positive attitudes towards "advertisement" and "awareness/awareness events" during this period.

Limitations and Future Work

  • Classification Accuracy: The automated classification of tweets into promotional and laypeople's discussions, while effective, is not perfect and may introduce some inaccuracies.
  • User Demographics: The study could not control for confounding variables like user demographics (age, gender, etc.) as this information is not readily available on Twitter.
  • Topic Granularity: The LDA model extracts abstract topics, and finer-grained aspects within topics (e.g., specific costs of genetic testing) were not explored in depth.
  • Representativeness of Twitter Users: Twitter users may not be representative of the general population, potentially skewing the findings.

Future work could involve analyzing user profiles and links within tweets for richer context, developing more advanced NLP tools, and exploring finer-grained topic analysis. The authors also suggest that awareness events might need to be more frequent to sustain their impact.

Summary updated

Abstract

There is evidence that participative social media platforms, namely Twitter, offer unique opportunities to inform cancer communication surveillance and to explore the mechanisms by which these new communication media affect individual health behavior and population health.
BibTeX
@article{bian2017using,
  author = {Jiang Bian and Yunpeng Zhao and Ramzi G. Salloum and Yi Guo and Mo Wang and Mattia Prosperi and Hansi Zhang and Xinsong Du and Laura J Ramirez-Diaz and Zhe He and Yuan Sun},
  title = {{Using Social Media Data to Understand the Impact of Promotional Information on Laypeople’s Discussions: A Case Study of Lynch Syndrome}},
  journal = {Journal of Medical Internet Research},
  year = {2017},
  doi = {10.2196/jmir.9266},
}
Figure