README.md

metadata

colorFrom: green
colorTo: green

⚔️ Game of Thrones Character Similarity Explorer

Explore semantic relationships between Game of Thrones characters using state-of-the-art embedding techniques. This project leverages SBERT and TF-IDF embeddings to map characters into high-dimensional semantic spaces, enabling similarity analysis, visualization, and recommendations.

---

🚀 Features

• Semantic Similarity Search: Find the most similar characters using cosine similarity on embeddings.
• Hybrid Embeddings: Combine SBERT (semantic) and TF-IDF (frequency-based) embeddings for improved results.
• Dimensionality Reduction Visualizations:
  • 2D & 3D t-SNE: Non-linear reduction highlighting local neighborhoods and clusters.
  • 2D & 3D PCA: Linear reduction emphasizing global variance and main directions.
• Interactive Plots: Explore embeddings with Plotly 2D/3D scatter plots, hoverable labels, and downloadable plots.
• Downloadable Results: Export similarity results as CSV for further analysis.

---

🧠 How It Works

Data Preprocessing

Character names are normalized, and embeddings are precomputed for SBERT and TF-IDF.

Similarity Computation

• SBERT embeddings capture contextual and semantic relationships.
• TF-IDF embeddings capture frequency-based textual features.
• Hybrid similarity is a weighted combination of both embeddings.

Recommendation

The app computes top-N similar characters based on the chosen embedding model and similarity type.

Visualization

• Reduce embeddings to 2D or 3D space with t-SNE or PCA.
• Interactive plots allow exploration of latent structures and clusters.

---

🛠️ Technologies Used

• Python & Streamlit for the web interface
• Pandas, NumPy, Joblib for data processing
• Scikit-learn for embeddings, similarity, t-SNE, and PCA
• Plotly for interactive 2D/3D visualizations
• Pillow for image handling
• SBERT & TF-IDF for semantic and frequency-based embeddings

---

⚙️ How to Use

1. Select a character from the sidebar.
2. Choose an embedding model: SBERT, TF-IDF, or Hybrid.
3. Adjust the number of similar characters to display.
4. Click Find Similar Characters to get results with images and similarity scores.
5. Explore Dimensionality Reduction to visualize embeddings in 2D/3D space.
6. Download CSV results or plots for offline use.

---

📊 Applications

• Character similarity analysis for literature and fandom studies
• Semantic search engines
• Recommendation systems
• Knowledge graph construction and clustering of entities

---

🔗 Live Demo

Game of Thrones Character Similarity Explorer

---

📄 License

MIT License – feel free to use and adapt for research or educational purposes.