Twitter Sentiment Analysis

Overview

This project performs sentiment analysis on Twitter data using a variety of machine learning and deep learning models, including traditional ML classifiers, LSTM, CNN, RNN, ANN, and transformer-based models (BERT, RoBERTa). The workflow covers data loading, cleaning, visualization, feature engineering, model training, evaluation, and comparison.

Features

• Data cleaning and preprocessing
• Exploratory data analysis and visualization (bar plots, histograms, word clouds)
• Feature engineering (TF-IDF, Word2Vec)
• Model training and evaluation:
  • Traditional ML: Logistic Regression, Random Forest, SVM, Naive Bayes, Decision Tree, KNN
  • Deep Learning: LSTM, CNN, RNN, ANN (using TensorFlow/Keras)
  • Transformer-based: BERT, RoBERTa (using HuggingFace Transformers)
• Hyperparameter tuning (GridSearchCV)
• Model comparison (accuracy, F1 score)

Dataset

• The notebook expects a CSV file containing Twitter data with columns such as Tweet, clean_tweet, sentiment, etc.
• The dataset is loaded interactively (e.g., via Google Colab's file upload) or by specifying a path.

Requirements

• Python 3.7+
• Jupyter Notebook or Google Colab
• Key libraries:
  • numpy, pandas, matplotlib, seaborn
  • scikit-learn
  • wordcloud, textblob
  • gensim
  • tensorflow, keras
  • torch, transformers (HuggingFace)

Install dependencies with:

pip install numpy pandas matplotlib seaborn scikit-learn wordcloud textblob gensim tensorflow torch transformers

Usage

1. Open the notebook: sentiment_analysis.ipynb in Jupyter or Google Colab.
2. Upload your dataset when prompted, or modify the code to load your CSV file directly.
3. Run all cells sequentially to:
   • Clean and explore the data
   • Visualize sentiment distributions and word clouds
   • Train and evaluate multiple models
   • Compare model performance

Models Implemented

• Traditional ML: Logistic Regression, Random Forest, SVM, Naive Bayes, Decision Tree, KNN
• Deep Learning: LSTM, CNN, RNN, ANN (TensorFlow/Keras)
• NLP Transformers: BERT, RoBERTa (HuggingFace Transformers)
• TextBlob: Rule-based sentiment analysis
• Word2Vec: Feature engineering for ML models

Results

• The notebook provides accuracy and F1 score comparisons for all models.
• Visualizations and tables summarize model performance.

Notes

• For BERT/RoBERTa, GPU acceleration is recommended (e.g., Google Colab).
• You may need to authenticate with HuggingFace for model downloads.
• Modify hyperparameters and model architectures as needed for your experiments.

License

This project is for educational and research purposes.