We focuses on the challenge of determining whether a given tweet indicates an actual disaster or simple exaggerated text. The model we use includes Transformers models(BERT, XLNet) and RNN models(LSTM and Bi-LSTM) together with multiple embedding strategies. We can achieve 84% F1 score in our validation data.
US 2020 turnout rate analysis using probability grpha model and recovered the dependency between turnout rate and the economic, political and pandemic factors.