Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Fearing fear itself - Exploring Fear Speech through Quantitative Analysis and Machine Learning: Sweden’s 2024 EU Election Campaigning on Facebook

von Schenck, Sofie LU (2025) SIMZ51 20251
Graduate School
Abstract
This thesis investigates the strategic use of fear speech, political communication
designed to evoke fear by presenting threats, by Swedish political parties and lead
candidates during the 2024 European Parliament election campaign on Facebook.
Departing from audience-focused studies, the analysis centres on political actors’
intent, examining who employs fear speech, how frequently, and on which topics.
A manually coded CamforS dataset of 803 posts was analysed, and a BERT-based
Swedish-language classification model was fine-tuned to detect fear speech.
Results show that fear speech is not confined to far-right or ideologically
extreme actors; both the Sweden Democrats and centre-left parties such as the
Social Democrats and the... (More)
This thesis investigates the strategic use of fear speech, political communication
designed to evoke fear by presenting threats, by Swedish political parties and lead
candidates during the 2024 European Parliament election campaign on Facebook.
Departing from audience-focused studies, the analysis centres on political actors’
intent, examining who employs fear speech, how frequently, and on which topics.
A manually coded CamforS dataset of 803 posts was analysed, and a BERT-based
Swedish-language classification model was fine-tuned to detect fear speech.
Results show that fear speech is not confined to far-right or ideologically
extreme actors; both the Sweden Democrats and centre-left parties such as the
Social Democrats and the Greens used it extensively, while the far-left employed it
least. Topic patterns aligned with party ideology.
The best-performing model achieved an F1-score of 0.78 on fear speech
detection. To address the inter-coder reliability challenges and conceptual
ambiguity of the term, the thesis proposes a reframing of “fear speech” as “threat
presentation”, focusing on observable rhetorical strategies rather than presumed
emotional effects. The findings contribute to the conceptual clarification of fear
speech, offer a methodological tool for its detection, and challenge prevailing
assumptions about its ideological distribution. (Less)
Please use this url to cite or link to this publication:
author
von Schenck, Sofie LU
supervisor
organization
course
SIMZ51 20251
year
type
H2 - Master's Degree (Two Years)
subject
keywords
fear speech, threat presentation, political communication, social media campaigning, machine learning classification
language
English
id
9209791
date added to LUP
2025-09-19 13:34:44
date last changed
2025-09-19 13:34:44
@misc{9209791,
  abstract     = {{This thesis investigates the strategic use of fear speech, political communication
designed to evoke fear by presenting threats, by Swedish political parties and lead
candidates during the 2024 European Parliament election campaign on Facebook.
Departing from audience-focused studies, the analysis centres on political actors’
intent, examining who employs fear speech, how frequently, and on which topics.
A manually coded CamforS dataset of 803 posts was analysed, and a BERT-based
Swedish-language classification model was fine-tuned to detect fear speech.
Results show that fear speech is not confined to far-right or ideologically
extreme actors; both the Sweden Democrats and centre-left parties such as the
Social Democrats and the Greens used it extensively, while the far-left employed it
least. Topic patterns aligned with party ideology.
The best-performing model achieved an F1-score of 0.78 on fear speech
detection. To address the inter-coder reliability challenges and conceptual
ambiguity of the term, the thesis proposes a reframing of “fear speech” as “threat
presentation”, focusing on observable rhetorical strategies rather than presumed
emotional effects. The findings contribute to the conceptual clarification of fear
speech, offer a methodological tool for its detection, and challenge prevailing
assumptions about its ideological distribution.}},
  author       = {{von Schenck, Sofie}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Fearing fear itself - Exploring Fear Speech through Quantitative Analysis and Machine Learning: Sweden’s 2024 EU Election Campaigning on Facebook}},
  year         = {{2025}},
}