Negativity spreads faster: A large-scale multilingual twitter analysis on the role of sentiment in political communication

Data and code accompanying the paper Negativity spreads faster: A large-scale multilingual twitter analysis on the role of sentiment in political communication.

In specific:

• the code used for the training of our models (./code/finetune_models.py and ./code/finetune_multi_cv.py)
• a Jupyter Notebook containing the major parts of our analysis (./code/analysis.ipynb)
• the model that was selected and used for the sentiment analysis.
• the manually annotated data used for training are shared (./data/annotation/).
• the ids of tweets that were used in our analyis and control experiments (./data/main/ & ./data/control)
• names, parties and handles of the MPs that were tracked (./data/mps_list.csv).

Annotated Data (./data/annotation/)

• One folder for each language (English, Spanish, Greek).
• In each directory there are three files:
  1. *_900.csv contains the 900 tweets that annotators labelled individually (300 tweets each annotator).
  2. *_tiebreak_100.csv contains the initial 100 tweets all annotators labelled. 'annotator_3' indicates the annotator that was used as a tiebreaker.
  3. *_combined.csv contains all tweets labelled for the language.

Model

Our model, xlm-twitter-politics-sentiment, along with a small tutorial on how to use it can be found in huggingface.co.

The model is based on the implementation of 'cardiffnlp/twitter-xlm-roberta-base-sentiment' while being further finetuned on the annotated dataset.

Example usage

from transformers import AutoModelForSequenceClassification, pipeline

model = AutoModelForSequenceClassification.from_pretrained('cardiffnlp/xlm-twitter-politics-sentiment')
sentiment_analysis_task = pipeline("sentiment-analysis", model=model, tokenizer="cardiffnlp/twitter-xlm-roberta-base-sentiment")

sentiment_analysis_task('Today is a good day')
Out: [{'label': 'Positive', 'score': 0.978614866733551}]

Reference paper

For more details, please check the reference paper. If you use the data contained in this repository for your research, please cite the paper using the following bib entry:

@article{antypas2023negativity,
  title={Negativity spreads faster: A large-scale multilingual twitter analysis on the role of sentiment in political communication},
  author={Antypas, Dimosthenis and Preece, Alun and Camacho-Collados, Jose},
  journal={Online Social Networks and Media},
  volume={33},
  pages={100242},
  year={2023},
  publisher={Elsevier}
}