Sentiment Analysis Using Random Forest and Naive Bayes

Overview

This project focuses on performing sentiment analysis on app reviews using two machine learning models: Random Forest and Naive Bayes. The goal is to classify user reviews into three sentiment categories: Positive, Neutral, and Negative. The project also includes a hybrid model that combines the predictions of both classifiers to improve accuracy.

Project Structure

Data Loading and Exploration: Load the dataset and perform an initial exploration to understand its structure. Data Preprocessing: Clean the data, handle missing values, and label sentiments based on the review scores. Exploratory Data Analysis (EDA): Visualize the data to gain insights into sentiment distribution, review length, and word frequency. Feature Extraction: Use the TF-IDF vectorizer to transform the text data into numerical features suitable for model training. Model Training: Train the Random Forest and Naive Bayes classifiers on the dataset. Model Evaluation: Evaluate the performance of both models using metrics such as accuracy, classification reports, and confusion matrices. Hybrid Model: Develop a hybrid algorithm that combines the predictions of both classifiers, incorporating confidence levels to refine the predictions. Model Comparison: Compare the accuracy and performance of the Random Forest, Naive Bayes, and Hybrid models. Findings: Analyze common sentiments, app popularity, and competitor comparisons.

Installation

To run this project, you'll need to install the following Python libraries:

pip install pandas numpy seaborn matplotlib scikit-learn wordcloud joblib

Usage

1. Load the Dataset: Ensure your dataset (app_reviews.csv) is placed in the specified directory.
2. Run the Code: Execute the script to load, preprocess, and analyze the data.
3. Train Models: The code will automatically train the Random Forest, Naive Bayes, and Hybrid models.
4. Evaluate Models: View the evaluation metrics and confusion matrices to assess model performance.
5. Test Hybrid Algorithm: Input your own review to see how the hybrid model classifies its sentiment.

Key Functions and Methods

label_sentiment(score): Labels the sentiment based on the review score. hybrid_algorithm(input_text): Combines Random Forest and Naive Bayes predictions to determine the sentiment with confidence levels. calculate_hybrid_accuracy(X_test, y_test): Computes the accuracy of the hybrid model on the test set. test_user_input_sentiment(input_text): Allows you to test the sentiment of custom review input using the hybrid model.

Results

Random Forest Accuracy: ~64% Naive Bayes Accuracy: ~58% Hybrid Model Accuracy: ~88%

The hybrid model outperformed individual classifiers by leveraging their strengths, resulting in higher accuracy and better sentiment prediction.

Visualization

Sentiment Distribution: Understand the distribution of sentiments across the dataset. Word Cloud: Visualize the most common words in reviews. Confusion Matrices: Evaluate model predictions against actual sentiments. Model Accuracy Comparison: Compare the accuracy of different models using bar charts.

Findings

Detailed insights into the most common sentiments, popular apps, and a comparison of competitor apps based on review sentiments and ratings.

Conclusion

This project demonstrates the effectiveness of combining multiple machine learning models to improve sentiment analysis. The hybrid model achieved a significant accuracy boost, making it a robust solution for analyzing user sentiments in app reviews.