兰德-Twitter对上疫苗误传的传播及对策（英文）-2021.4-81页_1mb

Summary of the Dissertation: Dissemination of Vaccine Misinformation on Twitter and Its Countermeasures

Core Content

This dissertation examines the dissemination of vaccine misinformation on Twitter, particularly in the context of the 2019 measles outbreak in the United States. It analyzes the production, spread, and consumption of vaccine-related misinformation, and proposes countermeasures for social media platforms, policymakers, and public health stakeholders.

Main Objectives

1. To develop a machine learning model for systematically collecting vaccine misinformation from Twitter.
2. To provide empirical evidence on the prevalence, composition, and trends of vaccine misinformation before and after the measles outbreak.
3. To characterize the diffusion network of vaccine misinformation using social network analysis.
4. To propose actionable strategies for combating vaccine misinformation and discuss the generalizability of the analytical framework.

Key Findings

Prevalence of Vaccine Misinformation

• Out of 1,721,528 vaccine-related tweets, 15% were not credible, 11% lacked evidence, and 18% were classified as propaganda.
• Vaccine misinformation primarily focuses on vaccine safety (65%), vaccine "truth" (39%), fraud and corruption (15%), conspiracy theories (14%), and autonomy (5%).
• Other categories include morality (5%), free speech, xenophobia, and efficacy (<3%).

Characteristics of Disseminators

• Most vaccine misinformation was spread by dedicated human actors rather than social bots.
• 15% of the misinformation disseminators were likely bots, and 4% were possibly completely automated.
• The top sources and propagators of misinformation often had vested interests in spreading false information to gain financial or social benefits.
• Social bots were more active in highly connected nodes (higher core numbers) of the retweet network, suggesting their strategic role in amplifying misinformation.

Role of Social Media and Algorithms

• Social media platforms and their algorithmic promotion of user-preferred content, combined with echo chamber effects, contribute to the spread of vaccine misinformation.
• User-generated content (blogs, podcasts, YouTube videos) and online interactions make the Internet a powerful tool for misinformation dissemination.

Countermeasures

• The dissertation proposes plausible actions for stakeholders, including:
  • Social media platforms: Implementing content moderation, suspending accounts, and adjusting search results.
  • Government and public health allies: Developing policies and educational campaigns to address vaccine hesitancy.
  • Domain experts and clinicians: Providing accurate, evidence-based information.
  • Web influencers: Using persuasive communication strategies to counter misinformation.
• The framework is generalizable to other forms of health misinformation and platforms beyond Twitter.

Methodology

Machine Learning Approach

• The study employed BERT (Bidirectional Encoder Representations from Transformers) and transfer learning to build a robust classifier for identifying vaccine misinformation.
• BERT was adapted from a pre-trained model to classify tweets based on their relevance and misinformation content.
• The model was trained on 3,972,651 tweets collected from January 2018 to April 2019, with a focus on tweets containing vaccine-related keywords and generated during the second week of each month.

Data Annotation

• A two-stage annotation process was used to ensure accuracy:
  • Stage 1: Classify tweets by relevance to vaccines (e.g., safety, effectiveness, conspiracy).
  • Stage 2: Classify relevant tweets as containing misinformation or not.
• Raters were required to correctly classify at least 80% of the test tweets to be qualified.
• HTML format was used to preserve tweet structure and media content during annotation.

Network Analysis

• Retweet network analysis and K-core decomposition were used to identify key influencers and the structure of misinformation spread.
• Social bots were found to be more active in highly connected parts of the network, suggesting their role in amplifying misinformation.
• The study highlights the need for interdisciplinary collaboration between social media platforms, policymakers, and public health experts to effectively counter misinformation.

Conclusion

The dissertation provides a comprehensive framework for understanding the spread of vaccine misinformation on Twitter. It emphasizes the importance of systematic data collection, machine learning applications, and social network analysis in identifying and mitigating the effects of vaccine misinformation. The proposed countermeasures are aimed at addressing the root causes of vaccine hesitancy and improving public trust in vaccines and medical institutions.