[Paper Review] Conspiracy in the Time of Corona: Automatic detection of Covid-19 Conspiracy Theories in Social Media and the News

Shadi Shahsavari, Pavan Holur+2 more|arXiv (Cornell University)|Apr 28, 2020

Misinformation and Its Impacts33 citations

TL;DR

The paper presents a pipeline to automatically detect and analyze narrative frameworks underlying Covid-19 conspiracy theories by extracting and linking actants and relationships from social media (4Chan, Reddit) and news (GDELT). It examines how these narratives align across platforms and evolve over time.

ABSTRACT

Rumors and conspiracy theories thrive in environments of low confidence and low trust. Consequently, it is not surprising that ones related to the Covid-19 pandemic are proliferating given the lack of any authoritative scientific consensus on the virus, its spread and containment, or on the long term social and economic ramifications of the pandemic. Among the stories currently circulating are ones suggesting that the 5G network activates the virus, that the pandemic is a hoax perpetrated by a global cabal, that the virus is a bio-weapon released deliberately by the Chinese, or that Bill Gates is using it as cover to launch a global surveillance regime. While some may be quick to dismiss these stories as having little impact on real-world behavior, recent events including the destruction of property, racially fueled attacks against Asian Americans, and demonstrations espousing resistance to public health orders countermand such conclusions. Inspired by narrative theory, we crawl social media sites and news reports and, through the application of automated machine-learning methods, discover the underlying narrative frameworks supporting the generation of these stories. We show how the various narrative frameworks fueling rumors and conspiracy theories rely on the alignment of otherwise disparate domains of knowledge, and consider how they attach to the broader reporting on the pandemic. These alignments and attachments, which can be monitored in near real-time, may be useful for identifying areas in the news that are particularly vulnerable to reinterpretation by conspiracy theorists. Understanding the dynamics of storytelling on social media and the narrative frameworks that provide the generative basis for these stories may also be helpful for devising methods to disrupt their spread.

Motivation & Objective

• Understand how Covid-19 conspiracy theories are generated using underlying narrative frameworks.
• Develop an automated pipeline to extract actants and relationships from social media and news sources.
• Quantify cross-domain information flow between social media and news about conspiracies during the pandemic.

Proposed method

• Model narratives as graphical networks with actants as nodes and labeled edges representing relationships.
• Extract relation tuples from sentences using dependency parsing and Semantic Role Labeling, and cluster noun phrases into contextual groups (CGs).
• Use BERT-based embeddings to cluster phrases into sub-nodes forming micro-contexts within the narrative graph.
• Aggregate social media posts into five-day segments and build co-occurrence actant networks for each segment.
• Compare social media-derived narratives with news reports using TF-IDF filtering and co-occurrence networks to study cross-corpus dynamics.
• Evaluate communities with coverage, homogeneity, completeness, and V-measure metrics across time.

Experimental results

Research questions

• RQ1 What narrative frameworks underlie Covid-19 conspiracy theories in social media and the news?
• RQ2 How do actants and their relationships align across social media and news over time?
• RQ3 Can an automated pipeline reveal the flow of conspiracy narratives from social platforms to news outlets?

Key findings

• The pipeline identifies five central phenomena in conspiracy narratives, including integration with existing theories, emergence of 5G-related theories, and alignment with anti-vaccine narratives.
• A daily-updated narrative graph reveals ebbs and flows in story interconnections and highlights news segments susceptible to conspiratorial reinterpretation.
• Social media communities have detectable presence in news reports, enabling a cross-corpus monitoring of information flow.
• The approach yields 52 social-media-derived communities with notable actants such as China, Trump, virus, Gates, and others driving narrative structures.
• The study demonstrates that narrative frameworks can be tracked and compared over time despite noisy social media data.