Sentiment Analysis Project

A comprehensive sentiment analysis system with seven different models: Multi-Layer Perceptron (MLP) Basic, MLP Enhanced, RoBERTa transformer-based, Kernel Approximation, Randomized PCA, LSTM, and LSTM-RoBERTa models. Includes a results tracking system that compares model performance.

Features

• Multiple sentiment analysis models:
  • Multi-Layer Perceptron (MLP) Basic: Simple neural network with TF-IDF features
  • Multi-Layer Perceptron (MLP) Enhanced: Improved neural network with advanced preprocessing
  • RoBERTa: Transformer-based model for state-of-the-art performance
  • Kernel Approximation: Approximate RBF kernel features with linear classification
  • Randomized PCA: Dimension reduction with logistic regression classifier
  • LSTM: Long Short-Term Memory neural network for sequence processing
  • LSTM-RoBERTa: Combined LSTM with RoBERTa embeddings for enhanced performance
• Performance tracking and comparison
• CSV and text output for model results

Getting Started

Prerequisites

• Python 3.8+
• TensorFlow 2.x
• NumPy, Pandas
• Scikit-learn
• PyTorch
• Transformers (Hugging Face)
• Matplotlib

Installation

pip install -r requirements.txt

Run the following inside the python shell:

>>> import nltk
>>> nltk.download()

Then an installation window appears. Go to the 'Models' tab and select 'punkt' & 'punkt_tab' from under the 'Identifier' column. Then click Download and it will install the necessary files.

Usage

Running the Pipeline

The main pipeline can be run with different modes:

python src/run_pipeline.py --mode [MODE] [OPTIONS]

Training Models

Train All Models

To train all models at once:

python src/run_pipeline.py --mode train_all --include_all --sample_size 5000

Train Individual Models

Train specific models using their dedicated modes:

# MLP Basic model
python src/run_pipeline.py --mode train_mlp_basic --sample_size 5000

# MLP Enhanced model
python src/run_pipeline.py --mode train_mlp_enhanced --sample_size 5000

# RoBERTa model
python src/run_pipeline.py --mode train_roberta --sample_size 1000

# Kernel Approximation model
python src/run_pipeline.py --mode train_kernel --sample_size 5000

# Randomized PCA model
python src/run_pipeline.py --mode train_pca --sample_size 5000

# LSTM model
python src/run_pipeline.py --mode train_lstm --sample_size 1000

# LSTM-RoBERTa model
python src/run_pipeline.py --mode train_lstm_roberta --sample_size 1000

Testing Models

Test All Models

To test all trained models at once:

python src/run_pipeline.py --mode test_all --include_all --sample_size 1000

Test Individual Models

Test a specific model using the test mode with the model name:

# Test MLP Basic model
python src/run_pipeline.py --mode test --test_model mlp_basic --sample_size 1000

# Test MLP Enhanced model
python src/run_pipeline.py --mode test --test_model mlp_enhanced --sample_size 1000

# Test RoBERTa model
python src/run_pipeline.py --mode test --test_model roberta --sample_size 1000

# Test Kernel Approximation model
python src/run_pipeline.py --mode test --test_model kernel --sample_size 1000

# Test Randomized PCA model
python src/run_pipeline.py --mode test --test_model pca --sample_size 1000

# Test LSTM model
python src/run_pipeline.py --mode test --test_model lstm --sample_size 1000

# Test LSTM-RoBERTa model
python src/run_pipeline.py --mode test --test_model lstm_roberta --sample_size 1000

Comparing Models

To compare all models without retraining:

python src/run_pipeline.py --mode compare --include_all --skip_training

Available Modes

Mode	Description
train_mlp_basic	Train the MLP Basic model
train_mlp_enhanced	Train the MLP Enhanced model
train_roberta	Train the RoBERTa model
train_kernel	Train the Kernel Approximation model
train_pca	Train the Randomized PCA model
train_lstm	Train the LSTM model
train_lstm_roberta	Train the LSTM-RoBERTa model
train_all	Train all models and compare them
test	Test a specific model
test_all	Test all models
compare	Compare existing trained models

Common Options

Option	Description	Default
--sample_size	Number of samples to use for training	20000
--verbose	Verbosity level (0=silent, 1=progress bar, 2=one line per epoch)	1
--test_model	Model to test when in 'test' mode (mlp_basic, mlp_enhanced, roberta, kernel, pca, lstm, lstm_roberta)	Required
--include_roberta	Include RoBERTa in the comparison (for 'compare' mode)	False
--include_all	Include all models in the comparison (for 'compare' mode)	False
--timeout	Timeout in seconds for each model script	1800
--skip_training	Skip model training and compare existing results	False

Individual Scripts

You can also run the individual scripts directly:

python src/scripts/mlp_basic_main.py --sample_size 10000
python src/scripts/mlp_enhanced_main.py --sample_size 10000
python src/scripts/roberta_main.py --sample_size 1000
python src/scripts/kernel_approximation_main.py --sample_size 10000
python src/scripts/randomized_pca_main.py --sample_size 10000
python src/scripts/lstm_main.py --mode train --sample_size 1000
python src/scripts/lstm_roberta_main.py --mode train --sample_size 1000
python src/scripts/compare_models.py --include_all --skip_training

Results

After running the models, results are stored in:

• src/model_results.json (raw data)
• src/model_results.csv (CSV format for easy loading into pandas)
• src/best_model_summary.txt (text summary of the best model)

The models themselves are stored in the src/models/ directory.