Tweet Sentiment Analysis

Goal

In this repository, we will experiment with various NLP approaches to tackle sentiment analysis.

Tasks Performed

• Analyse the tweet sentiment data
• Utilize VADERS method as a benchmark for our sentiment prediction model
• Perform sentence embedding using TFIDF vectorizer
• Experiment with Naïve Bayes and Linear Support Vector Classifier to predict tweet sentiment
• Architect a recurrent neural network model with large pretrained GLOVE word embeddings
• Finetune a BERT model to demonstrate the power of large pre-trained models

Overall Performance:

1. TDIDF with Naïve Bayes - 62.5%
2. TFIDF with Linear Support Vectir Classifier - 67.3%
3. GLOVE and LSTM - 69.7%
4. Finetuned BERT model - 79.3%