• Two million campaign-related tweets → sentiment classification with RNNs.
  • Bayesian popularity model infers candidate support from aggregated sentiment counts.