Twitter Sentiment Analysis with Streamlit & Docker

ML web app that can be deployed through Docker to analyze the sentiment of tweets. It uses Docker for the deployment and Streamlit for the web app.

The model used is a pre-trained model from Hugging Face. Concretely, it is a roBERTa model finetuned for sentiment analysis with the TweetEval benchmark. The model is available here.

Configuration

This app expects 2 untracked files to be present in the project:

• .streamlit/secrets.toml with the following content:

password = "your_password"

This file is used to protect the app with a password. You can change the password to whatever you want or even deactivate the if check_password in app.py to simply avoid this step. I prefer to have a password to avoid random people accessing the app in case I'd choose to test it on the open.

• .env with the following content:

API_KEY=YOUR_API_KEY
API_KEY_SECRET=YOUR_API_KEY_SECRET
BEARER_TOKEN=YOUR_BEARER_TOKEN
ACCESS_TOKEN=YOUR_ACCESS_TOKEN
ACCESS_TOKEN_SECRET=YOUR_ACCESS_TOKEN_SECRET

This file is used to authenticate the app to the Twitter API. You can get these credentials by creating a Twitter developer account and creating a new app. You can find more information here.

In case you just want to run the app for analyzing single tweets instead of user accounts, I have created a second prompt that does not require the Twitter API credentials as it depends solely on the model and the text you input.

How to run

A docker-compose.yml file is provided to run the app. You can run it with:

docker-compose up

This should have created the webapp for you to explore and access it at http://localhost:8501.

Changes should be reflected in the web app automatically thanks to the volumes of the docker-compose.yml file.

To stop the app, you can use Ctrl+C or docker-compose down in case you are running it on daemon mode.

References

• Streamlit
• HuggingFace
• TweetEval
• Tweepy
• Twitter API