aybukeatalay/uk-campaign-sentiment-roberta

UK Campaign Sentiment RoBERTa

This model is a fine-tuned version of cardiffnlp/twitter-roberta-base-sentiment for sentiment classification of tweets posted by UK general election candidates in the 2024 campaign period. It is part of a broader project introducing a multimodal dataset of campaign content, including text, images, and video.

Model Details

• Developed by: Christopher Barrie, Aybüke Atalay and Alia ElKattan
• Model type: RoBERTa-base (fine-tuned)
• Language: English
• Fine-tuned from: cardiffnlp/twitter-roberta-base-sentiment
• License: MIT

Training Details

• Training data: Manually annotated tweets from 2024 UK election candidates.
• Classes: Negative (−1), Neutral (0), Positive (1)
• Training period: 4 epochs with learning rate 2e−5 and batch size 8

Uses

This model is intended for sentiment analysis of political tweets, especially campaign-related content during UK elections. It can be applied to study negativity, campaign tone, or partisan differences in emotional framing.

Limitations

• The original model achieved approximately 72% accuracy on a manually annotated validation set, with strongest performance on neutral tweets.
• While this version has been fine-tuned on UK election campaign tweets, it may not generalize well to other domains or more informal, non-political language.

Citation

If you use this model, please cite:

Barrie, C., Atalay, A., & ElKattan, A. (upcoming). An enriched, multimodal social media dataset of a UK General Election campaign. [Dataset and model documentation in progress]

Contact

For questions or collaboration, contact Aybüke Atalay at: [email protected]