spChalk / Vaccine-Sentiment-Classifier

:syringe: Vaccine Sentiment Classifier is a deep learning classifier trained on real world twitter data, that distinguishes 3 types of tweets: Neutral, Anti-vax & Pro-vax.

💉 Vaccine-Sentiment-Classifier

Vaccine Sentiment Classifier (or VSC) is a Deep Learning classifier trained on real world twitter data.

VSC distinguishes 3 types of tweets:

1. 😐 Neutral
2. 😠 Anti-vax
3. ☺️ Pro-vax

The main aim of the project, is to showcase the multitude of ways (from shallow to deep learning) one can use NLP in order to extract sentiment from a given set.

Structure

The project is divided into 4 different implementations.

Each of them includes a .ipynb notebook and its corresponding documentation.

In a nutshell, the most important topics of each implementation are listed below.

1. VSC using Softmax Regression

• Data Cleaning, preprocessing, visualization
• n-grams
• TF-IDF
• Count vectorizer
• Hashing vectorizer
• Softmax regressor
• Hyperparameter tuning, using evolutionary algorithm: GASearchCV
• Learning curves & Classification report

2. VSC using Feed Forward Neural Networks (FFNN)

• Dealing with imbalanced data
• Optuna hyperparameter tuner
• FFNN using Bag of Words (BoW), Term Frequency–Inverse Document Frequency (TF-IDF), Word embeddings
• GloVe
• RoC curve

3. VSC using Recurrent Neural Networks (RNN)

• Bidirectional RNN
• LSTM / GRU
• Gradient clipping
• Skip Connections
• Attention mechanism

4. VSC using BERT

• Huggingface BERT moidels