Twitter Sentiment Analysis

Inspiration

• Train a good twitter sentiment detection model for the provided dataset.

What it does

• Detect and classify tweet sentiment into 4 classes: "Irrelevant", "Neutral", Positive", "Negative"

How we built it

• Using Kaggle Notebook for GPU resource
• Use Python and the Trainer API framework (from Hugging Face) together with PyTorch to train the model

Challenges we ran into

• Weak correlation between the validation score and the test score in the given dataset

Accomplishments that we're proud of

• A working model with an acceptable accuracy

What we learned

• Natural Language Processing concepts
• PyTorch and the transformers python library

What's next for Twitter Sentiment Analysis

• Faster and more accurate model
• A web application to use the model

Built With

• python
• pytorch
• transformers