Customer Review Sentiment Analysis using Naive Bayes

Overview

This project performs Sentiment Analysis on customer reviews using the Naive Bayes algorithm.
The goal is to classify text reviews into Positive or Negative sentiments based on the words used.

The model uses text preprocessing, TF-IDF vectorization, and Multinomial Naive Bayes to understand how words contribute to the sentiment of a review.

---

Libraries Used

• Python
• pandas – data loading and manipulation
• numpy – numerical operations
• scikit-learn (sklearn) – machine learning model, vectorization, metrics
• nltk – natural language preprocessing
• tqdm – progress tracking
• matplotlib & seaborn – data visualization
• BeautifulSoup & re – text cleaning (regex and HTML removal)

---

Project Workflow

1. Data Loading

• Load and inspect the dataset containing customer text reviews and sentiment labels.

2. Data Preprocessing

• Removed URLs, HTML tags, numbers, and non-alphabetic characters
• Converted all text to lowercase
• Tokenized and removed stopwords
• Lemmatized/stemmed words to their base form

3. Feature Extraction

• Used TF-IDF Vectorization to convert text into numerical features that represent word importance.

4. Model Building

• Trained a Multinomial Naive Bayes classifier using sklearn.naive_bayes.MultinomialNB.
• Optimized the smoothing parameter alpha using GridSearchCV.

5. Model Evaluation

• Evaluated performance using:
  • Accuracy
  • Precision, Recall, F1-score
  • ROC Curve and AUC Score
• Visualized top features contributing to Positive and Negative classes.

6. Results

• Displayed most informative features for each sentiment.
• Plotted Confusion Matrix, ROC Curve, and Word Importance charts.
• Model achieved high accuracy on unseen test data, demonstrating strong generalization.

---

How to Run

1. Clone this repository:

   git clone https://github.com/NirmalanSK/Customer-Review-NaiveBayes-SentimentAnalysis.git