🧠 SimpleNN - Sentiment Analysis Sandbox

Learning Playground

An educational, beginner-friendly sandbox designed to compare traditional Machine Learning and Deep Learning approaches for sentiment classification (Positive vs. Negative).

This repository serves as a hands-on learning project to understand the core differences between TF-IDF vectorization with Naive Bayes, and word embeddings with Neural Networks.

---

---

📂 Project Structure

• sentiment_classifier_nb.py: A clean, single-file Python script implementing Multinomial Naive Bayes with TF-IDF vectorization.
• sentiment_classifier_nn.ipynb: A comprehensive Jupyter Notebook containing the training pipeline for a TensorFlow Sequential Neural Network.
• sentiment_model.keras: The pre-trained, production-ready Deep Learning model exported from Keras.
• tokenizer.json: The saved tokenizer configuration containing the word-to-index index mapping for text preprocessing.

---

🛠️ Approaches

1. Traditional Machine Learning (sentiment_classifier_nb.py)

• Technology Stack: scikit-learn, numpy
• Methodology:
  • Uses TF-IDF Vectorization (Term Frequency-Inverse Document Frequency) to transform raw sentences into numerical matrices based on word relevance.
  • Implements a Multinomial Naive Bayes (MultinomialNB) classifier.
• Key Feature: Extremely fast training and evaluation times. Highly interpretable; allows inspection of class probabilities.

2. Deep Learning (sentiment_classifier_nn.ipynb)

• Technology Stack: tensorflow, keras, numpy
• Architecture:
  • Embedding Layer: Projects words into a 16-dimensional continuous vector space.
  • Global Average Pooling 1D: Flattens temporal sequence dimensions down to average representations.
  • Dense Layer (ReLU): Extracts high-level non-linear features (16 neurons).
  • Output Layer (Sigmoid): Outputs a continuous probability range between 0.0 (Highly Negative) and 1.0 (Highly Positive).
• Key Feature: Able to capture semantic meanings and word order sequences. Saves artifacts natively for easy integration into inference pipelines.

---

🚀 How to Run

Prerequisites

Make sure you have Python installed and the required dependencies set up:

pip install tensorflow scikit-learn numpy jupyter

Running the Naive Bayes Classifier

Run the standalone Python script directly from your terminal:

python sentiment_classifier_nb.py

Running the Neural Network Pipeline

1. Open the notebook in VS Code (with the Jupyter extension installed) or launch Jupyter Notebook:

   jupyter notebook

2. Open and execute the cells inside sentiment_classifier_nn.ipynb.
3. The notebook will automatically train the model and save sentiment_model.keras and tokenizer.json locally upon completion.

---

📊 Dataset Preview

Both classifiers are trained on a curated corpus of 100 annotated reviews (50 Positive / 50 Negative) representing real-world customer feedback variations:

• Positive Example: "Exceeded my expectations in every way"
• Negative Example: "Broke within the first five minutes of use"

---

🔮 Future Enhancements

• Expand the dataset to include multi-class emotions (Neutral, Angry, Excited).
• Integrate modern transformer models like DistilBERT for state-of-the-art accuracy.
• Build a lightweight web API using FastAPI to serve real-time inferences.

---

📜 License

This project is licensed under the MIT License. Feel free to use, modify, and distribute it! See the LICENSE file for more details.