UKElectionNarratives: A Dataset of Misleading Narratives Surrounding Recent UK General Elections

A collective of authors from the University of Sheffield, Athens University of Economics and Business, Piksel, and the Joint Research Centre, published a paper titled “UKElectionNarratives: A Dataset of Misleading Narratives Surrounding Recent UK General Elections.” The article introduces the first human-annotated dataset and taxonomy of misleading narratives circulating during the UK General Elections in 2019 and 2024.

authors describe the construction of UKElectionNarratives, which involved collecting, filtering, and manually annotating 2,000 tweets according to a multi-level codebook of 32 misleading narratives organized under 10 overarching super-narratives. The taxonomy covers a wide range of topics, including election integrity, economic uncertainty, social issues, foreign interference, and political polarization.

The paper also presents benchmarking experiments using pre-trained language models and large language models, including GPT-4o, to detect misleading election narratives. Results highlight the dataset’s effectiveness in training models to identify complex and nuanced political disinformation. The authors additionally emphasize the importance of combining human annotation with advanced language models to improve detection accuracy and reproducibility.

The article was Published under the Proceedings of the Nineteenth International AAAI Conference on Web and Social Media (ICWSM 2025) and can be accessed here.

‍