✨ Sentiment Analysis

Overview

This project demonstrates a machine learning pipeline for sentiment analysis using real-world Twitter data related to airline customer feedback. It is designed as a professional AI portfolio piece to showcase proficiency in natural language processing (NLP), data preprocessing, model training, and evaluation.

The ultimate goal is to develop a scalable sentiment analysis tool that can be extended to platforms such as WhatsApp messages, live chat systems, or customer support tools.

---

🎯 Project Goals

• Build a production-ready sentiment classifier using classical ML algorithms.
• Demonstrate applied knowledge of NLP and text preprocessing techniques.
• Use clean, real-world data from Kaggle to train and evaluate the model.
• Offer a visual explanation of model performance using classification metrics and confusion matrices.
• Prepare for future expansion with deep learning models or integration into web interfaces.

---

📁 Dataset

• Source: Kaggle - Airline Tweets Dataset
• Number of tweets: ~15,000
• Sentiment labels: positive, neutral, negative

---

⚙️ Workflow

1. Data Loading & Cleaning
   Load CSV data, remove noise, lowercase text, strip special characters.

2. Text Vectorization
   Apply CountVectorizer to convert text into numerical features.

3. Model Training
   Train a Multinomial Naive Bayes model using scikit-learn.

4. Model Evaluation
   Generate a classification report and confusion matrix to visualize model accuracy and class-wise performance.

---

📈 Model Performance

• Overall Accuracy: 78%
• Strong performance on the negative class (Precision: 0.78, Recall: 0.96)
• Moderate performance on the positive class (Precision: 0.82, Recall: 0.55)
• Lower performance on the neutral class (Precision: 0.72, Recall: 0.35) – affected by class imbalance

Classification Report Sample:

          precision    recall  f1-score   support

negative       0.78      0.96      0.86      1889
 neutral       0.72      0.35      0.47       580
positive       0.82      0.55      0.66       459

accuracy                           0.78      2928

macro avg 0.77 0.62 0.66 2928
weighted avg 0.77 0.78 0.75 2928

---

🔍 Example Prediction

model.predict(vectorizer.transform(["I love this airline!"]))
# Output: ['positive']

🧠 Future Plans

• Replace CountVectorizer with TF-IDF or deep embeddings like Word2Vec or BERT
• Try stronger models such as Logistic Regression, SVM, or deep learning classifiers
• Address class imbalance using SMOTE or class weighting
• Deploy the model as a simple web app using Streamlit or Gradio
• Extend to multilingual sentiment analysis (Arabic/English)
• Create a real-time API for processing WhatsApp or chat messages

🚀 Setup & Run

1. Install requirements

  pip install -r requirements.txt

2. Run the notebook Open Sentiment_Analysis_Project.ipynb in Jupyter Notebook or VS Code.

📦 File Structure

📁 Sentiment_Analysis
│
├── Sentiment_Analysis_Project.ipynb       # Main notebook
├── README.md                              # Project documentation
├── requirements.txt                       # Python dependencies
│
└── 📁 Data_Explorer
    ├── Tweets.csv                         # Raw dataset
    └── database.sqlite                    # SQLite version of dataset

---

📌 Author

Omar Khamis

AI & Robotics Enthusiast | Python Developer

• 💼 LinkedIn | 💻 GitHub
• 📧 Email: [email protected]

---

📜 License

This project is licensed under the Apache 2.0 License.