[Paper Review] Characterizing Online Engagement with Disinformation and Conspiracies in the 2020 U.S. Presidential Election

Karishma Sharma, Emilio Ferrara+1 more|arXiv (Cornell University)|Jul 17, 2021

Misinformation and Its Impacts26 references4 citations

TL;DR

This study analyzes 242 million election-related tweets from June–September 2020 to detect and characterize disinformation and conspiracy narratives, particularly QAnon-related content, using a trained CSI model (AUC 0.8, F1 0.76). It finds that while disinformation tweets are less viral than reliable content, QAnon accounts significantly engaged with left-leaning users and adapted to Twitter’s July 2020 ban by shifting hashtags and relying on pre-existing accounts, with statistically significant behavioral changes confirmed via regression discontinuity design.

ABSTRACT

Identifying and characterizing disinformation in political discourse on social media is critical to ensure the integrity of elections and democratic processes around the world. Persistent manipulation of social media has resulted in increased concerns regarding the 2020 U.S. Presidential Election, due to its potential to influence individual opinions and social dynamics. In this work, we focus on the identification of distorted facts, in the form of unreliable and conspiratorial narratives in election-related tweets, to characterize discourse manipulation prior to the election. We apply a detection model to separate factual from unreliable (or conspiratorial) claims analyzing a dataset of 242 million election-related tweets. The identified claims are used to investigate targeted topics of disinformation, and conspiracy groups, most notably the far-right QAnon conspiracy group. Further, we characterize account engagements with unreliable and conspiracy tweets, and with the QAnon conspiracy group, by political leaning and tweet types. Finally, using a regression discontinuity design, we investigate whether Twitter's actions to curb QAnon activity on the platform were effective, and how QAnon accounts adapt to Twitter's restrictions.

Motivation & Objective

• To identify and characterize disinformation and conspiracy narratives in U.S. 2020 election discourse on Twitter.
• To investigate the reach, engagement patterns, and propagation dynamics of unreliable and conspiracy-driven content, especially from QAnon.
• To evaluate the effectiveness of Twitter’s July 2020 ban on QAnon by analyzing behavioral adaptation and sustained engagement.
• To understand how political leaning, tweet types, and account activity influence engagement with disinformation and conspiracy content.

Proposed method

• Trained and applied the CSI (Cross-Source Inference) model to classify tweets as factual or unreliable/conspiratorial, using text, temporal, and account suspiciousness features.
• Used a dataset of 242 million election-related tweets collected between June 20 and September 6, 2020.
• Employed a regression discontinuity design (RDD) to estimate causal effects of Twitter’s QAnon ban, modeling hashtag usage changes around the intervention date.
• Compared propagation dynamics of unreliable/conspiracy and reliable tweet cascades using cascade breath and time-to-unique-accounts metrics.
• Validated model performance using AUC (0.8) and F1 (0.76) on a held-out test set.
• Analyzed engagement patterns by political leaning and tweet type, focusing on QAnon’s strategy of replying to left-leaning accounts to 'red pill' users.

Experimental results

Research questions

• RQ1 What are the prevalent disinformation and conspiracy narratives on Twitter preceding the U.S. 2020 Election?
• RQ2 How significant is the impact and reach of disinformation and conspiracy groups in terms of account engagements, political leaning, tweet types, and propagation dynamics?
• RQ3 Did Twitter’s restrictions on QAnon influence its activities and were they effective in limiting the conspiracy?

Key findings

• The CSI model achieved an AUC of 0.8 and F1 score of 0.76 in detecting unreliable and conspiratorial claims in election-related tweets.
• The most prominent disinformation topics included mail-in voter fraud, COVID-19, Black Lives Matter, media censorship, and false claims about political candidates.
• Disinformation tweets had lower viral reach than reliable tweets, with a higher mean time to reach unique accounts and smaller mean cascade breath.
• While 85.43% of all accounts (10.4M) showed no engagement with QAnon, this fraction dropped to 34.42% among the 1.2 million most active accounts.
• QAnon accounts engaged in targeted replies to left-leaning users to 'red pill' them, indicating a strategic effort to expand influence.
• Using regression discontinuity design, the study found statistically significant changes in hashtag adoption post-ban, with sustained engagement driven primarily by accounts created before the ban.