CeDA | Center for Data Analytics

Characterizing the interactions between politicians and voters in the era of social media

Keywords: Twitter, Natural Language Processing, Deep Learning, Graph theory
Project partner: Stefanie Bailer
CeDA collaborator: Sébastien Boyer

Project objectives

  • Characterizing the different flavors of interactions (positive, negative, offensive, etc...) between politicians and the general public on social media.

Approaches

  • Multilingual context dependent word embedding for sentences classification.
  • Graph representation of social media interaction.

Analytical methods

  • Transformer-based Deep Neural Networks: Multilingual BERT or XML-RoBERTa.
  • Graph Convolutional Neural Networks.

Figure 1. Tweets sentiment analysis based on context dependent word embedding, allows us to draw with confidence, directed edges representing emotional flow between politician tweets.