Political Elections Under (Social) Fire? Analysis and Detection of Propaganda on Twitter

Ansgar Kellner, Lisa Rangosch, Christian Wressnegger, Konrad RieckarXiv 2019 · arXiv

Introduction

In the wake of the 2016 US Presidential election, concerns regarding foreign interference and "fake news" became central to political discourse. Researchers from the Institute of System Security at Technische Universität Braunschweig investigated whether these tactics were mirrored during the 2017 German federal election (Bundestagswahl). By analyzing 9.5 million tweets, the study aimed to map online propaganda and develop robust tools for social bot detection.

Methodology

The researchers employed a three-pronged analytical framework:

  1. Actor Correlation: Cross-referencing 2,752 known IRA troll accounts from the US election with the German dataset.
  2. Landscape Analysis: Examining hashtags, media, and user interactions to identify amplified political agendas.
  3. Bot Detection: Engineering 44 unique features across metadata, textual content, temporal patterns, and user-based metrics to train a machine learning classifier.

The study utilized a refined corpus of 8.8 million tweets after filtering noise. A "gold standard" training set of 505 bots and 874 humans was manually labeled to facilitate supervised learning.

Troll Recycling and Temporal Activity

A significant finding was the "recycling" of state-sponsored infrastructure. The analysis identified 79 IRA trolls previously active in the US election who had pivoted to German politics. While their total tweet volume was relatively low (approx. 9,300 tweets), their presence underscores the persistent nature of cross-border propaganda.

Creation dates and activity levels of IRA troll accounts.

The temporal analysis revealed that tweet volume was highly sensitive to real-world political events. Spikes in activity coincided with party manifestos, state elections, and televised debates, culminating in a massive surge on Election Day. The data shows that retweets significantly outnumbered original tweets during these peak periods, suggesting a high degree of content amplification.

Temporal distribution of tweets and retweets relative to key political events in 2017.

Key Insights: AfD Dominance and Bot Prevalence

The right-wing Alternative für Deutschland (AfD) dominated the digital landscape. The hashtag #afd appeared nearly 2 million times, tripling the volume of the second-place party, the SPD. However, high volume did not always equate to support; much of the viral content included satire and protest messages, reflecting the controversial nature of the discourse.

The researchers estimated that approximately 12.19% of the most active users were bots. This aligns with international trends, suggesting that the German Twittersphere is similarly susceptible to automation.

Bot Detection Performance

The study evaluated seven machine learning models, with GradientBoosting emerging as the most effective. It achieved an Area Under Curve (AUC) of 0.976 and an F1-score of 0.891. Notably, this custom classifier significantly outperformed the industry-standard Botometer (AUC 0.802), highlighting the importance of localizing detection tools to specific political and linguistic contexts.

Conclusion

The research demonstrates that state-sponsored actors and automated bots played a measurable role in the 2017 German election. While right-wing narratives were most heavily amplified, the study also reveals the complexity of distinguishing propaganda from organic controversy. Future work will integrate Natural Language Processing (NLP) to better understand the specific narratives pushed by these automated actors.

Summary updated

Abstract

For many, social networks have become the primary source of news, although the correctness of the provided information and its trustworthiness are often unclear. The investigations of the 2016 US presidential elections have brought the existence of external campaigns to light aiming at affecting the general political public opinion. In this paper, we investigate whether a similar influence on political elections can be observed in Europe as well. To this end, we use the past German federal election as an indicator and inspect the propaganda on Twitter, based on data from a period of 268 days. We find that 79 trolls from the US campaign have also acted upon the German federal election spreading right-wing views. Moreover, we develop a detector for finding automated behavior that enables us to identify 2,414 previously unknown bots.
BibTeX
@misc{kellner2019political,
  author = {Ansgar Kellner and Lisa Rangosch and Christian Wressnegger and Konrad Rieck},
  title = {{Political Elections Under (Social) Fire? Analysis and Detection of Propaganda on Twitter}},
  year = {2019},
  eprint = {1912.04143},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1912.04143},
}
Figure