KarinaAulia/Movie_Review_Summarizer

🎬 Smart Text Summarizer: TextRank vs T5-small

<div align="center">

Domain-aware text summarization system comparing Extractive and Abstractive approaches

Try Demo • Report Bug • Documentation

</div>

🌟 Overview

This project implements and compares two fundamental text summarization approaches:

• —TextRank (Extractive): Graph-based algorithm that extracts key sentences from the original text
• —T5-small (Abstractive): Transformer-based model that generates new paraphrased summaries

Key Features

✨ Real-time ROUGE Scoring - Automatically calculates quality metrics for every input 🎯 Domain Detection - Intelligently identifies movie reviews vs general text 📊 Three Comparison Modes - Multiple perspectives on summary quality 🚀 Universal Application - Works on any text type, optimized for movie reviews ⚡ Fast Processing - Instant results with efficient algorithms 🎨 Interactive UI - Beautiful, user-friendly Gradio interface

🎯 Demo

Try different text types to see how both models perform:

1. 1.Movie Reviews (Positive/Negative) - Primary domain
2. 2.News Articles - General news content
3. 3.Academic Papers - Technical/scientific text
4. 4.Stories & Narratives - Creative writing
5. 5.Your Own Text - Paste anything!

🔬 Methodology

TextRank (Extractive Summarization)

Original Text → Sentence Tokenization → Similarity Matrix → 
PageRank Algorithm → Ranked Sentences → Extract Top-K → Summary

Algorithm:

• —Graph-based ranking using PageRank
• —Sentence similarity via word overlap
• —Preserves original phrasing
• —Fast and efficient

Pros:

• —✅ Grammatically correct (uses original sentences)
• —✅ Factually accurate
• —✅ Fast processing
• —✅ No training required

Cons:

• —❌ Less fluent connections
• —❌ May miss context
• —❌ Limited paraphrasing

T5-small (Abstractive Summarization)

Original Text → Tokenization → T5 Encoder → 
Transformer Layers → T5 Decoder → Generated Summary

Model:

• —Pre-trained T5-small (60M parameters)
• —Text-to-Text Transfer Transformer
• —Beam search decoding (num_beams=4)
• —Max length: 150 tokens

Pros:

• —✅ More fluent and natural
• —✅ Better paraphrasing
• —✅ Captures abstract concepts
• —✅ Flexible output

Cons:

• —❌ May introduce errors
• —❌ Slower processing
• —❌ Requires more resources

📊 Real-time ROUGE Calculation

This app calculates ROUGE scores in real-time for every input!

Three Comparison Modes:

1️⃣ Summaries vs Original Text

Measures content retention and overlap

TextRank vs Original → Shows extractive quality
T5 vs Original → Shows abstractive quality

2️⃣ Model Agreement (TextRank vs T5)

Shows consensus between approaches

High Agreement → Both identify similar key points
Low Agreement → Different focus/perspectives

3️⃣ Interpretation Guide

📈 ROUGE Metrics Explained

• —ROUGE-1: Unigram (single word) overlap
• —ROUGE-2: Bigram (two consecutive words) overlap
• —ROUGE-L: Longest Common Subsequence

🎓 Domain Detection

Smart algorithm that identifies text type:

Detection Logic

Strong Keywords: film, movie, director, actor, screenplay
Medium Keywords: plot, character, scene, performance
Weak Keywords: good, bad, great, entertaining

Score = (Strong × 3) + (Medium × 2) + (Weak × 1)

Confidence Levels

📚 Evaluation Results

Dataset Evaluation (Colab)

Evaluated on 100 movie reviews with ground truth summaries:

Key Findings:

• —✅ T5 outperforms TextRank by 21.4% in ROUGE-1
• —✅ T5 shows 44.4% improvement in ROUGE-2 (better phrase capture)
• —✅ TextRank is ~10x faster than T5
• —✅ TextRank better for factual accuracy, T5 for fluency

Performance Metrics

🛠️ Technical Stack

Core Libraries

transformers==4.35.0      # T5 model
torch==2.1.0              # PyTorch backend
nltk==3.8.1               # Text processing
networkx==3.2.1           # Graph algorithms
rouge-score==0.1.2        # Evaluation metrics
gradio==4.7.1             # Web interface
scikit-learn==1.3.2       # ML utilities

System Requirements

• —Python: 3.8+
• —RAM: 2GB minimum (4GB recommended)
• —GPU: Optional (CPU works fine)
• —Storage: ~500MB for models

🚀 Usage Examples

Example 1: Movie Review

Input:

"This film is a masterpiece of modern cinema. The director's vision is crystal clear throughout..."

TextRank Output:

"This film is a masterpiece of modern cinema. The performances are outstanding. The soundtrack complements the narrative perfectly."

T5 Output:

"the film is a masterpiece with breathtaking cinematography and outstanding performances. while pacing may feel slow, it serves the story well."

ROUGE Scores:

• —TextRank: R-1: 0.45, R-2: 0.22, R-L: 0.38
• —T5: R-1: 0.52, R-2: 0.28, R-L: 0.45

Example 2: News Article

Input:

"The technology sector experienced significant volatility today..."

TextRank Output:

"The technology sector experienced significant volatility. Share prices fluctuated dramatically. Investors remain cautious but optimistic."

T5 Output:

"major companies announced quarterly earnings, causing significant volatility in the technology sector amid supply chain challenges."

🎯 Use Cases

Educational

• —📚 Teaching NLP concepts
• —🎓 Comparing summarization approaches
• —🔬 Research demonstrations

Practical

• —📰 News article summarization
• —🎬 Movie review digests
• —📄 Document processing
• —📧 Email summarization

Development

• —🧪 Testing summarization models
• —📊 Benchmarking algorithms
• —🔄 Comparing extractive vs abstractive

🔍 How It Works

Step-by-Step Process

1. 1.Input Processing

   User Input → Domain Detection → Preprocessing

1. 1.Parallel Summarization

   TextRank: Text → Sentences → Graph → PageRank → Summary
   T5: Text → Tokenize → Encode → Generate → Summary

1. 1.Real-time Evaluation

   ROUGE Calculation → Statistics → Display Results

1. 1.Interactive Display

   Summaries + Metrics + Visualizations → User Interface

📊 Statistics Provided

For each summary, you get:

Summary Metrics

• —Word count (original vs summary)
• —Sentence count
• —Character count
• —Compression ratio (%)

ROUGE Scores

• —ROUGE-1 (unigram overlap)
• —ROUGE-2 (bigram overlap)
• —ROUGE-L (longest common subsequence)

Model Info

• —Processing method
• —Algorithm type
• —Speed/performance
• —Quality indicators

🎨 Interface Features

Clean & Modern Design

• —🎨 Purple gradient theme
• —📱 Responsive layout
• —🌓 Professional styling
• —✨ Smooth animations

User-Friendly

• —🎯 Clear instructions
• —📋 Sample texts included
• —🔄 One-click testing
• —📊 Visual comparisons

Interactive Elements

• —✅ Real-time processing
• —📈 Live statistics
• —🎬 Domain indicators
• —🗑️ Easy reset

🧪 Testing

Try these scenarios to test the system:

1. 1.Long Reviews (200+ words) - Test compression
2. 2.Short Reviews (50 words) - Test edge cases
3. 3.Technical Text - Test domain detection
4. 4.Mixed Content - Test robustness
5. 5.Multiple Languages - Test limitations (English only)

📖 Academic Context

Project Information

Course: Natural Language Processing Topic: Text Summarization Comparison Approach: Extractive vs Abstractive Evaluation: ROUGE Metrics

Learning Objectives

✅ Understanding graph-based algorithms (TextRank) ✅ Working with transformer models (T5) ✅ Implementing evaluation metrics (ROUGE) ✅ Comparing summarization approaches ✅ Building interactive NLP applications

Research Questions

1. 1.How do extractive and abstractive methods compare?
2. 2.When is each approach more suitable?
3. 3.What are the trade-offs in speed vs quality?
4. 4.How does domain affect summarization quality?

🤝 Contributing

This is an educational project, but suggestions are welcome!

How to Contribute

1. 1.Fork the repository
2. 2.Create your feature branch
3. 3.Test your changes
4. 4.Submit a pull request

Areas for Improvement

• —[ ] Add more domain detection categories
• —[ ] Implement fine-tuned T5 for movie reviews
• —[ ] Add more evaluation metrics
• —[ ] Support for other languages
• —[ ] Batch processing capability

📝 Citation

If you use this project in your research or education, please cite:

@software{smart_text_summarizer,
  title={Smart Text Summarizer: TextRank vs T5-small},
  author={Your Name},
  year={2024},
  url={https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE}
}

🔗 References

Papers

• —TextRank: Mihalcea & Tarau (2004) - "TextRank: Bringing Order into Text"
• —T5: Raffel et al. (2020) - "Exploring the Limits of Transfer Learning"
• —ROUGE: Lin (2004) - "ROUGE: A Package for Automatic Evaluation of Summaries"

Resources

• —Hugging Face Transformers
• —NetworkX Documentation
• —NLTK Documentation
• —Gradio Documentation

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

MIT License

Copyright (c) 2024

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction...

👤 Author

Your Name

• —GitHub: @yourusername
• —LinkedIn: Your Name
• —Email: your.email@example.com

🙏 Acknowledgments

• —Hugging Face for hosting and transformers library
• —Google for T5 pre-trained model
• —NLTK community for text processing tools
• —Gradio team for the amazing interface framework
• —Course Instructor for guidance and support

📞 Support

Having issues or questions?

• —📧 Email: your.email@example.com
• —💬 Issues: GitHub Issues
• —📚 Documentation: Wiki

🗺️ Roadmap

Version 1.0 ✅

• —[x] TextRank implementation
• —[x] T5 integration
• —[x] Real-time ROUGE scoring
• —[x] Domain detection
• —[x] Gradio interface

Version 2.0 (Planned)

• —[ ] Fine-tuned T5 for movie reviews
• —[ ] Multi-language support
• —[ ] Batch processing
• —[ ] Advanced metrics (BERTScore)
• —[ ] Export functionality

<div align="center">

⭐ Star this project if you find it helpful!

🎬 Made with ❤️ for NLP Education

⬆ Back to Top

</div>