Fig. 3
Schematic to understand the logic of sentiment index generation. Upper part shows the steps to train sentiment models and select the one with the best performance. After designing the structure of a neural classifier, we train the model using 80% of the Sentiment 140 dataset and evaluate the performance of the remaining 20%. All models are trained on the same training dataset and evaluated on the same test dataset to ensure consistency. Lower part describes the steps to generate a local sentiment index. We input the vectorized geotagged tweets into the selected best model. Then, we assign a value between 0 and 1 to represent the sentiment score for each tweet using the SoftMax function. Finally, we aggregate the tweet level sentiment score to county/city, state, or country level to represent the local sentiment index.
