Twitter is a social media network on which users post and interact with messages known as "tweets". It allows a user to post, like, and retweet tweets.
Twitter is also known for the excessive negativity or criticism by many of its users. Considering that, this application intends to classify a tweet according to its positivity.
The application uses a pre-trained BERT model fine tuned using the Coronavirus tweets NLP dataset. This dataset contains 48.000 tweets classified into the following five categories:
- Extremely negative
- Negative
- Neutral
- Positive
- Extremely positive
Currently, the application is deployed using AWS, accessible in this URL.
The following diagram represents the CD pipeline, which is currently hosted in GitHub Actions.
On the other hand, the following diagram represents an usual inference workflow. The user writes the tweet's url into the Gradio frontend, the tweet's text is scrapped and the lambda function invoked.
- Implement a CI pipeline
- Implement a CD pipeline
- Implement a CT pipeline
- Deploy the backend on AWS Lambda
- Deploy the gradio app on an EC2 instance
- Implement a monitoring solution to detect data drift