Umeå University's logo

umu.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Detection of Synthetic Climate Misinformation with Machine Learning Algorithms and Sentence-Level Analysis
Umeå University, Faculty of Science and Technology, Department of Computing Science.
2025 (English)Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesis
Abstract [en]

The spread of climate-related misinformation can reduce public support for climate change mitigation policies. A study showed that on social media, people tend to absorb news content without knowing the details of the context. In that case, LLM can be utilised to spread misinformation, subsequently altering people's opinions for malicious purposes. To observe two machine learning algorithms: Support Vector Machine and Logistic Regression's capability to detect LLM-generated misinformation, we created a synthetic dataset, consisting of 300 examples. We have collected 150 climate-related news articles from various well-reputed sources to create the synthetic dataset. Then, we created a five to six-sentence summary based on the original article with the help of GPT-4. Each actual summary is falsified with the help of GPT-4 as well. Moreover, we evaluated each summary example from the synthetic dataset with the FineSure framework to obtain each summary's faithfulness, completeness and conciseness. The results showed that Support Vector Machine achieved an F1-score of 0.839, and Logistic Regression's F1-score was 0.787 on the synthetic dataset. We performed sentence-level analysis with the GUTEK framework on these models' false positive and negative examples. The sentence-level analysis with the GUTEK framework showed that policy-related sentences had the most impact on these models in predicting false positives. On the other hand, factual-related sentences significantly influenced these models to predict false negatives. 

Place, publisher, year, edition, pages
2025.
Series
UMNAD ; 1576
National Category
Computer Sciences
Identifiers
URN: urn:nbn:se:umu:diva-242908OAI: oai:DiVA.org:umu-242908DiVA, id: diva2:1988038
Educational program
Bachelor of Science Programme in Computing Science
Examiners
Available from: 2025-08-11 Created: 2025-08-10 Last updated: 2025-08-11Bibliographically approved

Open Access in DiVA

fulltext(2976 kB)146 downloads
File information
File name FULLTEXT01.pdfFile size 2976 kBChecksum SHA-512
79d718865e25d5c5c7a8ee9afdb3c79c53e54f20d7b8b8d04a91d7f305f87ac7478e352468ad12e1055c8f3330c40f5df7f95edb6986856a53439830bb4fefdf
Type fulltextMimetype application/pdf

By organisation
Department of Computing Science
Computer Sciences

Search outside of DiVA

GoogleGoogle Scholar
Total: 146 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 757 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf