🎬 Movie Recommendation System (Hybrid ML + Web App)

A full-stack movie recommendation system that combines keyword-based similarity and semantic similarity using modern NLP techniques, deployed as a web application using Flask.

This project demonstrates the end-to-end lifecycle of an ML system — from data preprocessing and feature engineering to model inference and web deployment.

---

🚀 Features

• Hybrid recommendation approach:

  • Bag-of-Words (CountVectorizer + Cosine Similarity)
  • Semantic similarity using Sentence-BERT

• Fuzzy matching for robust movie name input

• Clean and minimal web interface

• Fast inference using precomputed vectors and embeddings

• Modular and production-oriented code structure

---

🧠 Recommendation Strategy

The system computes movie similarity using a weighted hybrid score:

Final Score = 0.6 × BoW Similarity + 0.4 × Semantic Similarity

This balances:

• Lexical overlap (genres, keywords, cast, crew)
• Contextual meaning (semantic understanding of movie descriptions)

---

🗂️ Dataset

The project uses multiple CSV files derived from the TMDB movie dataset, including:

• Movie metadata
• Cast and crew information
• Keywords and genres
• Poster paths

Since these files are not row-aligned, an ID-based data integration pipeline was implemented to ensure correctness.

⚠️ Raw datasets are excluded due to size and licensing constraints. See data/README.md for preprocessing details.

---

🏗️ Project Structure

movie-recommender/
│
├── data/
│   ├── processed_movies.csv
│   ├── recommendations.json
│   └── README.md
│
├── notebooks/
│   └── data_preprocessing.ipynb
│
├── models/
│   ├── vectors.npz
│   ├── sbert_embeddings.npy
│   ├── stopwords.pkl
│   └── sbert_model/
│
├── templates/
│   └── index.html
│
├── static/
│   └── script.js
│
├── app.py
├── recommender.py
├── requirements.txt
├── README.md
└── .gitignore

---

⚙️ Tech Stack

• Python
• Pandas, NumPy
• Scikit-learn
• Sentence-Transformers (SBERT)
• PyTorch
• Flask
• HTML, CSS, JavaScript

---

⚡ Performance Optimizations

• Offline preprocessing and feature extraction
• Cached BoW vectors and SBERT embeddings
• Models and data loaded once at server startup
• No recomputation during user requests

This design ensures low latency and scalability for web deployment.

---

▶️ How to Run Locally

1️⃣ Clone the repository

git clone https://github.com/your-username/movie-recommender.git
cd movie-recommender

2️⃣ Create and activate environment

conda create -n movie-recommender python=3.10
conda activate movie-recommender

3️⃣ Install dependencies

pip install -r requirements.txt

4️⃣ Run the application

python app.py

Visit:

http://127.0.0.1:5000

---

🧪 Example Output

[
  {
    "title": "Interstellar",
    "poster": "https://image.tmdb.org/t/p/w500/xyz.jpg",
    "score": 0.87
  }
]

---

📌 Key Learnings

• Importance of ID-based data alignment
• Hybrid recommendation system design
• ML model deployment pitfalls and fixes
• Flask + ML integration best practices
• Performance optimization for inference-time systems

---

🔮 Future Improvements

• User-based collaborative filtering
• Search auto-completion
• Explanation for recommendations
• Cloud deployment (Docker / Render)
• User feedback loop

---

👤 Author

Harshavardhan B.Tech CSE, NIT Trichy Interests: NLP, Recommendation Systems, ML Systems, Applied AI

---

⭐ Acknowledgements

• TMDB for the dataset
• Sentence-Transformers library
• Open-source ML community