Source-linked AI summary
Who Let The Trolls Out? Towards Understanding State-Sponsored Trolls
Savvas Zannettou, Tristan Caulfield, William Setzer, Michael Sirivianos, Gianluca Stringhini, Jeremy Blackburn
TL;DR
State-sponsored trolls’ operations, content, evolving strategies, and Web-wide influence remain poorly understood. The paper analyzes Russian and Iranian troll activity using Twitter and Reddit datasets, content and temporal analyses, and Hawkes Processes to estimate cross-platform influence. It finds ideological and temporal variation, real-world-event-linked campaigns, and greater Russian influence on most studied platforms except /pol/.
Problem
Research had not thoroughly established how state-sponsored trolls operate, what they disseminate, how their behavior changes, or how much they influence the Web.
Method
The paper analyzes Twitter and Reddit troll datasets across behavior, content, targets, and time, using word embeddings and Hawkes Processes for cross-platform influence estimation.
Results
Russian trolls were pro-Trump and more influential and efficient than Iranian trolls on Twitter, Reddit, and Gab, while Iranian trolls were more influential on /pol/.
Takeaways & Limitations
Changing troll behavior makes automated detection difficult, while the measured influence extends beyond the originating platforms into other Web communities.
Takeaways & Limitations
The 4chan dataset begins in late June 2016, and Gab launched in August 2016, constraining temporal coverage for those platforms.
Abstract
from arXiv · showhide
Recent evidence has emerged linking coordinated campaigns by state-sponsored actors to manipulate public opinion on the Web. Campaigns revolving around major political events are enacted via mission-focused "trolls." While trolls are involved in spreading disinformation on social media, there is little understanding of how they operate, what type of content they disseminate, how their strategies evolve over time, and how they influence the Web's information ecosystem. In this paper, we begin to address this gap by analyzing 10M posts by 5.5K Twitter and Reddit users identified as Russian and Iranian state-sponsored trolls. We compare the behavior of each group of state-sponsored trolls with a focus on how their strategies change over time, the different campaigns they embark on, and differences between the trolls operated by Russia and Iran. Among other things, we find: 1) that Russian trolls were pro-Trump while Iranian trolls were anti-Trump; 2) evidence that campaigns undertaken by such actors are influenced by real-world events; and 3) that the behavior of such actors is not consistent over time, hence automated detection is not a straightforward task. Using the Hawkes Processes statistical model, we quantify the influence these accounts have on pushing URLs on four social platforms: Twitter, Reddit, 4chan's Politically Incorrect board (/pol/), and Gab. In general, Russian trolls were more influential and efficient in pushing URLs to all the other platforms with the exception of /pol/ where Iranians were more influential. Finally, we release our data and source code to ensure the reproducibility of our results and to encourage other researchers to work on understanding other emerging kinds of state-sponsored troll accounts on Twitter.
1 Introduction
The paper studies how Russian and Iranian state-sponsored trolls operate, what they disseminate, how their behavior changes, and how they influence the Web. Using Twitter and Reddit data, it finds ideological, temporal, campaign, and cross-platform differences between the groups.
- The study addresses limited understanding of how state-sponsored trolls operate, what they disseminate, how their behavior changes, and how much they influence the Web.
- The authors analyze 10M tweets from Russian and Iranian trolls and posts from 944 Russian trolls on Reddit across behavior, evolution, targets, and shared content.
- Russian trolls were more influential and efficient than Iranian trolls on Twitter, Reddit, and Gab, but not on /pol/.
- Russian trolls were pro-Trump, whereas Iranian trolls were anti-Trump.
- Iranian campaigns against France and Saudi Arabia coincided with real-world events affecting those countries’ relations with Iran.
- Troll behavior changed over time, including substantial shifts in language and Twitter-client use, complicating automated detection.
- Russian trolls discussed cryptocurrencies extensively on Reddit, unlike Russian trolls on Twitter to the same extent, and the authors released source code for reproducibility.
2 Related Work
Prior work examined opinion manipulation, political disinformation, and bots, but this study focuses on ground-truth Russian and Iranian trolls and quantifies their influence across multiple Web communities.
- Earlier studies examined sockpuppets, forum manipulation, campaign detection, and politically motivated disinformation.
- Related research also analyzed bot activity, political agendas, and information spreading during events such as Brexit and the Russia-Ukraine conflict.
- Unlike previous work, this study uses Russian and Iranian troll accounts independently identified and suspended by Twitter and Reddit.
- The paper extends prior work by quantifying troll influence on Twitter, Reddit, 4chan, and Gab.
3 Background
The paper studies Twitter, Reddit, 4chan, and Gab as influential Web communities, using 4chan and Gab specifically for cross-platform influence estimation.
- The study selects Twitter, Reddit, 4chan, and Gab because they are impactful actors in the Web’s information ecosystem.
- Twitter: Twitter is a mainstream network where users broadcast short tweets containing features such as hashtags and mentions.
- Reddit: Reddit organizes URL posts into topic-focused subreddits whose popularity and ordering are shaped by votes.
- 4chan: 4chan is an anonymous imageboard organized into topical boards, and the study focuses on its Politically Incorrect board, /pol/.
- Gab: Gab combines Twitter-like broadcasting with Reddit-like voting and has extremely lax moderation policies.
4 Troll Datasets
The datasets combine publicly released Twitter and Reddit accounts identified as state-sponsored trolls with their posts and account information. The paper documents dataset construction and notes ethical and coverage constraints.
- The study uses datasets of Russian and Iranian trolls on Twitter and Reddit.
- Twitter’s released dataset contains accounts identified as state-sponsored trolls, with an assumed near absence of false positives but unknown false negatives.
- Reddit provided 944 accounts identified as operated on behalf of the Russian government.
- The account-creation data are summarized in Figure 1, while Table 2 reports the top ten words and bigrams in Twitter troll profile descriptions.
- Reddit submissions, comments, and account details were recovered from Pushshift dumps and crawled user pages, whose union contains gaps from removals and deletions.
- The analysis uses publicly available data, follows standard ethical guidelines, and does not attempt to de-anonymize users.
5 Analysis
The analysis finds that Russian and Iranian troll campaigns were event-linked, targeted different audiences, and changed their account activity, language, and posting strategies over time.
- Account creation: 80% of Russian troll accounts were created before 2016, while Iranian accounts had larger creation bursts after the 2016 US presidential election.Russian account creation also peaked around the Republican National Convention and the Unite the Right rally; 75% of Russian Reddit accounts were created in the first half of 2015.
- Account characteristics: 25% of Iranian trolls had more than 1k followers versus 15% of Russian trolls, with median follower counts of 392 and 132, respectively.Iranian trolls also had a lower median followers-to-friends ratio: 0.51 versus 0.74 for Russian trolls.
- Temporal activity: Russian and Iranian troll activity formed distinct temporal patterns tied to campaigns, including Ukrainian-conflict activity, Trump-related Reddit spikes, and Iranian activity concentrated from early morning to 13:00 UTC.Russian Twitter activity followed a diurnal pattern, while Russian Reddit and Iranian Twitter activity decreased during the weekend.
- Campaign evolution: Russian troll campaigns shifted from a diverse Ukraine-targeting account set to later campaigns, while mention and retweet strategies changed across periods.Russian trolls used this strategy particularly in 2014–2015, whereas Iranian trolls began using it after August 2017.
- Languages and clients: Russian and Iranian trolls used language and clients differently over time, with Russian trolls initially posting mostly in Russian and Iranian campaigns shifting toward French and English.Russian trolls began with near-exclusive use of the twitterfeed client, while Iranian trolls initially relied almost exclusively on the facebook Twitter client.
6 Influence Estimation
The paper estimates how URLs shared by Russian and Iranian trolls spread across Twitter, Reddit, Gab, and /pol/ using Hawkes Processes. Russian trolls generally had greater influence and efficiency across communities, except on /pol/, while The Donald showed high external influence.
- Data and approach: The analysis tracks URLs shared by Russian trolls, Iranian trolls, or both across Reddit, Twitter, Gab, and 4chan’s /pol/.The datasets cover Reddit and Twitter from January 2016 to October 2018, /pol/ from July 2016 to October 2018, and Gab from its available period.
- Data and approach: Hawkes Processes estimate how likely URL events in one community are to cause subsequent events in another.The models are fit separately for each unique URL to estimate connection strengths between communities.
- Metrics: Absolute influence measures the percentage of destination events caused by a source community, while influence relative to size measures URL-pushing efficiency.The relative-to-size metric is the number of destination events caused as a percentage of source-platform events.
- Absolute influence: Russian trolls were more influential in spreading URLs to other communities overall, except on /pol/, where Iranian trolls were more influential.Russian trolls were particularly influential to Gab (1.9%), the rest of Twitter (1.29%), and /pol/ (1.08%).
- Efficiency: Russian trolls were most efficient at pushing URLs to Twitter (10.4%) and Gab (3.19%), whereas Iranian trolls were most efficient on Twitter (3.6%) and the rest of Reddit (2.04%).For combined Russian- and Iranian-shared URLs, Russian trolls were more efficient across communities except /pol/, where Iranians were more efficient.
- Cross-community influence: The Donald had the highest external influence on other platforms for both Russian- and Iranian-troll URL groups.The paper identifies The Donald as an impactful actor in the information ecosystem and a possible vector for pushing information to other communities.
7 Discussion & Conclusion
The paper concludes that Russian and Iranian trolls differ in their targets, content, and platform-specific behavior, while their tactics also change over time. These patterns make automated identification difficult and show meaningful influence in fringe online communities.
- Findings: Russian and Iranian trolls differed according to origin and operating platform, including differences in campaigns, behavioral evolution, and disseminated content.The study also found platform-specific differences such as cryptocurrency discussions among Russian trolls on Reddit but not to the same extent on Twitter.
- Implications: Troll tactics and talking points overlap with regular-user behavior, while changes over time make automated identification and blocking an open challenge.The authors state that a classifier effective for one campaign might not detect future campaigns.
- Implications: State-sponsored trolls had meaningful influence in fringe communities including The Donald, 4chan’s /pol/, and Gab, where pushed topics resonated strongly.The authors connect this pattern to users sympathetic to the trolls’ views or to unidentified state-sponsored actors in those communities.