Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Detecting Anti-vaccine Content on Twitter using Multiple Message-Based Network Representations

James R. Ashford

TL;DR

This study investigates detecting anti-vaccine content on Twitter by constructing three message-based interaction networks (mentions, replies, and quote retweets) for each term related to COVID-19. It ranks terms via crowdsourced Likert ratings, hydrates tweets to obtain content and metadata, and extracts both global and local graph features from 597 term–interaction networks to train binary classifiers. Global network features, particularly when using a random forest classifier and combining all interaction types, yield the strongest performance (about 0.886 accuracy), while local features offer competitive but generally lower gains. The findings demonstrate that language-agnostic, graph-based representations can effectively distinguish controversial from non-controversial terms at scale, with potential applications in content moderation across platforms and languages.

Abstract

Social media platforms such as Twitter have a fundamental role in facilitating the spread and discussion of ideas online through the concept of retweeting and replying. However, these features also contribute to the spread of mis/disinformation during the vaccine rollout of the COVID-19 pandemic. Using COVID-19 vaccines as a case study, we analyse multiple social network representation derived from three message-based interactions on Twitter (quote retweets, mentions and replies) based upon a set of known anti-vax hashtags and keywords. Each network represents a certain hashtag or keyword which were labelled as "controversial" and "non-controversial" according to a small group of participants. For each network, we extract a combination of global and local network-based metrics which are used as feature vectors for binary classification. Our results suggest that it is possible to detect controversial from non-controversial terms with high accuracy using simple network-based metrics. Furthermore, these results demonstrate the potential of network representations as language-agnostic models for detecting mis/disinformation at scale, irrespective of content and across multiple social media platforms.

Detecting Anti-vaccine Content on Twitter using Multiple Message-Based Network Representations

TL;DR

This study investigates detecting anti-vaccine content on Twitter by constructing three message-based interaction networks (mentions, replies, and quote retweets) for each term related to COVID-19. It ranks terms via crowdsourced Likert ratings, hydrates tweets to obtain content and metadata, and extracts both global and local graph features from 597 term–interaction networks to train binary classifiers. Global network features, particularly when using a random forest classifier and combining all interaction types, yield the strongest performance (about 0.886 accuracy), while local features offer competitive but generally lower gains. The findings demonstrate that language-agnostic, graph-based representations can effectively distinguish controversial from non-controversial terms at scale, with potential applications in content moderation across platforms and languages.

Abstract

Social media platforms such as Twitter have a fundamental role in facilitating the spread and discussion of ideas online through the concept of retweeting and replying. However, these features also contribute to the spread of mis/disinformation during the vaccine rollout of the COVID-19 pandemic. Using COVID-19 vaccines as a case study, we analyse multiple social network representation derived from three message-based interactions on Twitter (quote retweets, mentions and replies) based upon a set of known anti-vax hashtags and keywords. Each network represents a certain hashtag or keyword which were labelled as "controversial" and "non-controversial" according to a small group of participants. For each network, we extract a combination of global and local network-based metrics which are used as feature vectors for binary classification. Our results suggest that it is possible to detect controversial from non-controversial terms with high accuracy using simple network-based metrics. Furthermore, these results demonstrate the potential of network representations as language-agnostic models for detecting mis/disinformation at scale, irrespective of content and across multiple social media platforms.
Paper Structure (20 sections, 2 equations, 7 figures, 5 tables)

This paper contains 20 sections, 2 equations, 7 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Dataset
  4. Methodology
  5. Ranking of Terms
  6. Hydrating tweets
  7. Network generation
  8. Network features
  9. Global network features
  10. Local network features
  11. Results
  12. Discovery of anti-vax terms
  13. Data Overview
  14. Global network features
  15. Local network features
  16. ...and 5 more sections

Figures (7)

  • Figure 1: Examples of three message-based interaction networks in the form of mentions (left), quote retweets (centre) and replies (right) on Twitter based on tweets which mention #NoNewNormal.
  • Figure 2: Distribution of mean score for all terms in the set with the threshold $t=0.95$ marked in red
  • Figure 3: Two-dimensional principal component analysis for all global network features combining reply, mention and quote retweet interactions
  • Figure 4: The corresponding eigenvector values for each principal component as shown in Figure \ref{['fig:pca_global_tweets']}
  • Figure 5: Two-dimensional principal component analysis for all local network features for each interaction type
  • ...and 2 more figures