In this project, we import data from https://www.kaggle.com/gpreda/covid19-tweets and train a few machine learning and deep learning models to do a sentiment analysis on labeled COVID-19 tweets. Then, we deploy these models to a web app on Streamlit so we can write our own tweets about COVID-19 and see our tweet's sentiments according to our trained models.
Raw tweets are preprocessed first to remove links, punctuations, and stop words. Then, it is vectorized into words, and lemmatized. Afterwards, using Tf-idf, each tweet is converted into vectors of numbers which are then processed with ML Algorithms. There are 4 machine learning model used: Logistic Regression, XGBoost, Light GBM, and Decision Tree.
For the deep learning model, we use embedding layer from TensorFlow before we create and train a model with Bidirectional LSTM layers. The result shows that the deep learning model achieves higher accuracy compared to our machine learning models.
Finally, we deploy the trained models on a web app on streamlit in which users can type their own 'tweets' about COVID-19 and let our models classify the sentiment.
To run the web app locally, you can download this repository and type streamlit run streamlit-sentiment.py in the folder's terminal.
To view the hosted web app, you can visit the following link: https://share.streamlit.io/radvian/covid-sentiment/main/streamlit-sentiment.py