Source-linked AI summary
Analyzing the Digital Traces of Political Manipulation: The 2016 Russian Interference Twitter Campaign
Adam Badawy, Emilio Ferrara, Kristina Lerman
TL;DR
The paper investigates how Russian-linked Twitter accounts affected the spread of misinformation before the 2016 U.S. election, focusing on ideological, bot, and geographic patterns among users who reshared their content. Using a large election-period dataset, network-based ideology labeling, bot detection, and geospatial and text analyses, it finds that conservative users especially amplified mostly conservative, pro-Trump troll content. The authors caution that the analysis cannot account for all malicious manipulation efforts during the election.
Problem
The paper asks whether political ideology, bot activity, and geography shaped engagement with and propagation of misinformation on Twitter.
Method
The study analyzes election-period Twitter data, uses label propagation to infer ideology, applies bot detection, and examines troll content and user geography.
Results
Conservatives shared Russian troll content more widely than liberals, while troll messages were mostly conservative and pro-Trump; bot activity appeared among both ideological groups.
Takeaways & Limitations
Although troll content reached users across the political spectrum, conservative Twitter accounts were especially concentrated in its consumption and dissemination.
Takeaways & Limitations
The analysis cannot account for all malicious efforts aimed at manipulation during the 2016 presidential election.
Abstract
from arXiv · showhide
Until recently, social media was seen to promote democratic discourse on social and political issues. However, this powerful communication platform has come under scrutiny for allowing hostile actors to exploit online discussions in an attempt to manipulate public opinion. A case in point is the ongoing U.S. Congress' investigation of Russian interference in the 2016 U.S. election campaign, with Russia accused of using trolls (malicious accounts created to manipulate) and bots to spread misinformation and politically biased information. In this study, we explore the effects of this manipulation campaign, taking a closer look at users who re-shared the posts produced on Twitter by the Russian troll accounts publicly disclosed by U.S. Congress investigation. We collected a dataset with over 43 million election-related posts shared on Twitter between September 16 and October 21, 2016, by about 5.7 million distinct users. This dataset included accounts associated with the identified Russian trolls. We use label propagation to infer the ideology of all users based on the news sources they shared. This method enables us to classify a large number of users as liberal or conservative with precision and recall above 90%. Conservatives retweeted Russian trolls about 31 times more often than liberals and produced 36x more tweets. Additionally, most retweets of troll content originated from two Southern states: Tennessee and Texas. Using state-of-the-art bot detection techniques, we estimated that about 4.9% and 6.2% of liberal and conservative users respectively were bots. Text analysis on the content shared by trolls reveals that they had a mostly conservative, pro-Trump agenda. Although an ideologically broad swath of Twitter users was exposed to Russian Trolls in the period leading up to the 2016 U.S. Presidential election, it was mainly conservatives who helped amplify their message.
1 INTRODUCTION
Social media can support democratic participation but also enable scalable misinformation and political manipulation. This paper examines who amplified Russian troll content before the 2016 U.S. election, along ideological, bot-related, and geographic dimensions.
- Motivation: Social media supported democratic conversation and political coordination, but also became a potential channel for manipulating opinion and spreading misinformation.Prior studies documented both democratizing uses and risks associated with deceptive political content.
- Research setting: 2,752 Twitter accounts tied to Russia’s Internet Research Agency were used as a proxy for fake news because investigators identified malicious intent.The study analyzes how messages from these now-deactivated accounts spread on Twitter.
- Research questions: The paper asks whether ideology, bots, and geography shaped engagement with and propagation of misinformation.The questions compare liberal and conservative participation, bot prevalence, and variation across U.S. states.
- Approach: 43 million election-related tweets from about 5.7 million users were collected, and user ideology was inferred with label propagation at precision and recall above 90%.The study used a retweet network and media-sharing behavior to classify users as liberal or conservative.
- Main findings: Russian trolls mostly promoted conservative causes and pro-Trump material, while conservatives disproportionately amplified their messages.The analysis also examined bot activity and the geographic distribution of participating users.
2 DATA COLLECTION
The study assembled a large election-period Twitter dataset, classified users through partisan media-sharing patterns, and identified Russian troll accounts for spread analysis. It then characterized retweet activity, media links, and the resulting retweet network.
- Twitter Dataset: 43.7 million unique tweets from nearly 5.7 million users were collected through continuous Twitter Search API queries during the 2016 election period.Queries used manually compiled election-related hashtags and keywords, with collection infrastructure deployed on AWS.
- Classification of Media Outlets: The media-source lists contained 190 liberal and 167 conservative outlets after cross-referencing with the Twitter dataset.Five most frequently appearing outlets from each partisan category were used to compile labeled user lists.
- Classification of Media Outlets: Users were classified as liberal or conservative according to whether they shared more links to liberal or conservative media sources.Users with equal numbers of links from both sides were removed from the final labeled set.
- Russian Trolls: 2,752 Twitter accounts identified by the U.S. Congress as Russian trolls were used to study troll activity in the collected data.221 appeared in the dataset, and 85 produced original tweets; trolls also retweeted other users and one another.
3 DATA ANALYSIS & METHODS
The analysis constructs a large retweet network, infers user ideology through label propagation, detects bots with Botometer, and maps activity using user-reported or geotagged locations.
- Retweet Network: 4,474,044 nodes comprise the retweet network’s giant component.Nodes represent Twitter users, with directed links when one user retweets another.
- Ideology Inference: Label propagation assigns ideology using network connections to liberal or conservative seed users.The seeds were derived from users who mainly retweeted partisan media outlets.
- Validation: Around 0.91 precision and recall were achieved in five-fold validation on the seed users.A second validation using hyper-partisan users produced scores around 0.93.
- Bot Detection: Botometer estimates whether accounts are bots using more than one thousand content, network, temporal, profile, and sentiment features.The analysis applied the Python API to inspect active users; accounts scoring above 0.5 were labeled bots.
- Geo-location: Self-reported profile locations map substantially more tweets than exact geolocation, but they contain fictitious or imprecise entries.Only 36,351 tweets had exact coordinates, whereas profile locations mapped over 10.5 million tweets to U.S. states.
4 RESULTS
The results show that conservative users substantially amplified Russian troll content more than liberal users, with activity concentrated in prominent Republican states and bot prevalence similar across ideological groups.
- 4.1 Activity of Russian Trolls: 107 liberal and 108 conservative Russian troll accounts were identified, but conservative trolls produced 844 original tweets versus 44 from liberal trolls.The groups were nearly balanced in account count but sharply different in original-content activity.
- 4.2 Users Engaged with Russian Trolls: 28,274 spreaders wrote original tweets, producing over 1.5 million original tweets and over 12 million total tweets and retweets.Spreaders are users who retweeted Russian trolls, excluding the trolls’ own activity from these totals.
- 4.2.1 Political Ideology: 892 liberal spreaders produced more than 42,000 tweets, while 27,382 conservative spreaders produced more than 1.5 million.Liberal spreaders’ content indicated support for Clinton, whereas conservative spreaders’ content openly supported Trump and linked to conservative outlets.
- 4.2.2 Geospatial Analysis: Conservative spreaders’ tweets came from more states, with Texas and Florida standing out as the biggest geographic sources, while New York stood out among liberal sources.The broader geographic spread among conservatives may be an artifact of their higher tweet volume.
- 4.2.3 Bots: 2,126 of 34,160 spreaders with bot scores exceeded 0.5 and were considered bots.Bot scores were available for 34,160 of the 40,224 spreaders.
- 4.2.3 Bots: Bot shares were 4.9% among liberal spreaders and 6.2% among conservative spreaders, while bot-score densities were similar and most users ranged from 0 to 0.6.The comparison is affected by the disproportionate number of conservative spreaders.
5 CONCLUSIONS
The paper uses Russian troll content as a proxy for misinformation to examine its spread during the 2016 election. It finds that conservatives shared this content more widely than liberals, while noting that the analysis cannot capture all manipulation efforts.
- 5 CONCLUSIONS: The study investigates misinformation by analyzing Twitter content produced by Russian trolls during the period leading to the 2016 U.S. presidential election.Tweets were collected from September 16 to October 21, 2016, using the Twitter Search API and manually compiled keywords and hashtags.
- 5 CONCLUSIONS: Conservatives shared Russian troll misinformation more widely than liberals, with about 30 times more conservative users and a 35:1 content-production ratio.The paper also reports that conservative trolls produced almost 20 times more content and outnumbered liberal trolls by about four to one.
- 5 CONCLUSIONS: Bot estimates were 4.9% for liberal users and 6.2% for conservative users.These estimates concern users who retweeted Russian trolls.
- 5 CONCLUSIONS: Misinformation spread by malicious actors can enhance malicious information and polarize political conversations, causing confusion and social instability.The paper identifies these as severe negative consequences of malicious misinformation spread.
- 5 CONCLUSIONS: The analysis cannot account for all malicious efforts aimed at manipulating the 2016 presidential election.The paper notes that state and non-state actors could deploy misinformation campaigns and social bots, motivating further research and improved detection techniques.