Abstract:
In this work, we used datasets on emotional responses and comments to COVID-19 to analyze people’s perceptions towards COVID-19. Based on the trend of perceptions, we predicted the trend for the next month.
Data:
Ground Truth: used for training the sentiment classification models.
Dataset used for prediction from January 1st to April 17th: used for topic extraction and prediction.
Models:
- Models used for sentiment classification:
Model Name | Accuracy |
---|---|
Naive Bayes | 0.8440 |
Logistic Regression | 0.9528 |
Random Forests | 0.7337 |
BERT | 0.8538 |
XLNet | 0.6547 |
Liear SVM | 0.7989 |
Linear SVC | 0.9218 |
LSTM | 0.9534 |
RoBERTa | 0.7410 |
DistilBERT | 0.8987 |
- Model used for topics extraction:
LDA
- Models used for sequential prediction:
Model Name | RMSE (Sentiment) | RMSE (Topic) |
---|---|---|
LSTMLB | 64.56 | 125.73 |
LSTMWM | 56.71 | 170.96 |
LSTMTS | 67.62 | 159.57 |
LSTMM | 277.72 | 242.68 |
ARIMA | 27.07 | 85.71 |
Details: