Punishiment or prevention?: A machine learning analysis
of drug-related news coverage across cultures

Research Highlights
− Analyzed 1,736 articles from South Korea and U.S. media
− Used BERT model to predict public engagement (70% accuracy)
− Revealed distinct cultural differences in drug-related news coverage

Key Methods
− Computational text analysis
− Machine learning prediction models
− Cross-cultural comparative analysis

Abstract

This study examines contrasting media framings of drug-related issues in South Korea and the United States (U.S.), emphasizing the need for culturally informed reporting that balances enforcement with health-oriented narratives. Using computational text analysis and machine learning (ML), 1,736 articles were analyzed to identify thematic differences, drivers of public engagement, and predictive models of audience interest.Results reveal South Korean media's predominant focus on enforcement and punitive measures, while U.S. media prioritize health and prevention. Public engagement in South Korea is driven by sensational elements, such as specific incidents and punitive measures. ML models, particularly BERT, achieved the highest accuracy in predicting audience responses, underscoring the potential of advanced computational approaches. This study offers actionable insights for ethical journalism, advocating for culturally nuanced, life-centered narratives that promote resolution and healing.
Keywords: Computational journalism, Machine learning, Media framing, News coverage, Public engagement, Public health communication

I am currently updating the code used in this research on this page. For inquiries, please reach out at reason[at]ewhain.com

Codes

The following is the process for Research Question 2 (including data collection and preprocessing.

  1. #1 Data collection through crawling
  2. #2 Data preprocessing  
  3. #3 Feature representation, selection, and visualization
  4. #3-1 Content-Comment Count Partitioning & Text Feature Comparison
  5. #3-2 Title-Comment Count Feature Comparison (+Comparison between Groups with Comments and without Comments)
  6. #4 Verification of 5 features