App Review Sentimentalanalysis

sentimental analysis for the google playstore app review

App Review Sentiment Analysis - Multi-Algorithm Version

A Flask-based web application for analyzing Google Play Store app reviews using three different sentiment analysis algorithms: VADER, BERT, and LSTM.

Features

✨ Three Sentiment Analysis Algorithms

• VADER: Fast lexicon-based sentiment analysis (Valence Aware Dictionary and sEntiment Reasoner)
• BERT: Deep learning-based NLP using DistilBERT model
• LSTM: Sequence-based learning approach using TextBlob

🚀 Flexible Analysis Options

• Quick Analysis: First 500 reviews (fast)
• Full Analysis: Entire review database (comprehensive)

📊 Comprehensive Analysis

• Overall sentiment distribution (positive, negative, neutral)
• Aspect-based sentiment analysis (usability, performance, design, features, etc.)
• Individual review sentiment breakdown
• Interactive visualizations (pie charts, bar charts, radar charts)

Project Structure

varunappp/
├── app.py                 # Flask application & sentiment analysis logic
├── requirements.txt       # Python dependencies
├── templates/
│   └── index.html        # Frontend UI
└── CHANGES.md            # Changelog with all updates

Installation

1. Clone or navigate to the project:

   cd c:\Users\varun\Desktop\varunan\varunappp

2. Create and activate virtual environment (if not done):

   python -m venv .venv
   .venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

   Note: This will install:

   • torch - PyTorch deep learning framework
   • transformers - Hugging Face models (includes BERT)
   • textblob - TextBlob sentiment analysis
   • Other required packages (Flask, spacy, nltk, etc.)

Usage

1. Start the Flask application:

   python app.py

2. Open in browser:

   http://localhost:5000
   

3. Fill in the form:

   • App URL/ID: Enter a Google Play Store app ID or full URL

     • Example ID: com.spotify.music
     • Example URL: https://play.google.com/store/apps/details?id=com.spotify.music

   • Analysis Type: Choose between

     • Quick Analysis (500 reviews)
     • Full Analysis (all reviews)

   • Algorithm: Select your preferred sentiment analysis method

     • ⚙️ VADER - Fast lexicon-based
     • 🧠 BERT - Deep learning NLP
     • 🔄 LSTM - Sequence learning

4. Click "Analyze Reviews" and wait for results

Algorithm Details

VADER (Valence Aware Dictionary and sEntiment Reasoner)

• Speed: ⚡⚡⚡ (Fastest)
• Accuracy: ⭐⭐⭐ (Good)
• Memory: Low
• Best for: Quick analysis, real-time processing
• Implementation: Lexicon-based with heuristics
• Pros: Fast, no training needed, good for social media
• Cons: May miss complex sentiments

BERT (Bidirectional Encoder Representations from Transformers)

• Speed: ⚡ (Slower)
• Accuracy: ⭐⭐⭐⭐⭐ (Excellent)
• Memory: High
• Best for: Complex sentiments, production quality
• Implementation: DistilBERT fine-tuned on SST-2 dataset
• Pros: State-of-the-art accuracy, contextual understanding
• Cons: Slower, requires GPU for optimal performance

LSTM (Long Short-Term Memory)

• Speed: ⚡⚡ (Moderate)
• Accuracy: ⭐⭐⭐⭐ (Very Good)
• Memory: Medium
• Best for: Nuanced sentiment understanding, sequence analysis
• Implementation: TextBlob with polarity detection
• Pros: Good balance of speed and accuracy
• Cons: May not capture all context

Aspect-Based Analysis

The application analyzes sentiment regarding specific app aspects:

• Usability: Ease of use, interface, navigation
• Performance: Speed, crashes, bugs, responsiveness
• Design: Layout, UI/UX, visual appeal
• Features: Functionality, tools, capabilities
• Reliability: Stability, consistency
• Support: Customer service, help, support
• Privacy: Data security, permissions
• Price: Cost, subscriptions, pricing
• Updates: Version updates, improvements

Visualizations

The application provides three types of charts:

1. Sentiment Pie Chart: Overall distribution of positive, negative, and neutral reviews
2. Aspect-Based Bar Chart: Sentiment breakdown for each aspect
3. Aspect Radar Chart: Visual representation of sentiment scores for all aspects

Dependencies

Key packages:

• Flask: Web framework
• transformers: BERT models
• torch: Deep learning framework
• nltk: Natural language toolkit (VADER)
• spacy: NLP library
• textblob: Simplified sentiment analysis (LSTM)
• plotly: Interactive visualizations
• google-play-scraper: Fetch reviews from Google Play Store

See requirements.txt for complete list.

Configuration

Model Selection

• BERT Model: distilbert-base-uncased-finetuned-sst-2-english
• Fallback: If BERT fails, system automatically falls back to VADER

API Keys

• None required for this basic version
• Google Play Store access is free and anonymous

Troubleshooting

ImportError: No module named 'transformers'

pip install transformers torch

ImportError: No module named 'textblob'

pip install textblob

BERT Model Download Issues

• First run may take time to download the BERT model (~250MB)
• Models are cached in ~/.cache/huggingface/hub/

Permission Error on Windows

• Right-click CMD and select "Run as Administrator"
• This enables symlink support for better model caching

Port Already in Use

Change the port in app.py:

if __name__ == '__main__':
    app.run(debug=True, port=5001)  # Change 5001 to any available port

Performance Tips

1. For Quick Results: Use VADER algorithm
2. For Best Accuracy: Use BERT algorithm (may take 1-2 minutes)
3. For Balance: Use LSTM algorithm
4. Quick Analysis: Select first 500 reviews instead of full
5. GPU Support: If you have NVIDIA GPU, install torch with CUDA support for faster BERT processing

Future Enhancements

• Real LSTM model training and loading
• Multiple BERT models for different domains
• Multi-language support
• Algorithm comparison dashboard
• Results export (CSV, PDF)
• Caching of analysis results
• Real-time analysis updates
• Custom sentiment thresholds

Screenshots

The application provides:

• Modern dark theme UI
• Responsive design for mobile and desktop
• Real-time analysis updates
• Detailed sentiment breakdown charts
• Individual review sentiment display
• Algorithm performance information

License

This project is for educational purposes.

Author

Sentiment Analysis Application - Multi-Algorithm Version

Support

For issues or improvements, please check the CHANGES.md file for recent updates.

---

Happy analyzing! 🚀