🎬 Netflix Content-Based Recommendation System

A modular machine learning system that recommends similar movies and shows using TF-IDF vectorization and cosine similarity on textual features.

---

🚀 Overview

This project implements a content-based filtering recommendation system that suggests titles similar to a given input, based on:

• Title — highest weight
• Genre — medium weight
• Description — base weight

Text features are combined, vectorized with TF-IDF, and compared using cosine similarity to rank and return the top-N most similar titles.

---

🧠 How It Works

Raw Data → Feature Engineering → TF-IDF Vectorization → Cosine Similarity → Top-N Recommendations

Step	Description
1. Load Data	Reads cleaned_data.csv into a pandas DataFrame
2. Feature Engineering	Combines title, genre, and description into a single weighted combined_text field
3. TF-IDF Vectorization	Converts text to numerical vectors; builds vocabulary and computes IDF scores
4. Cosine Similarity	Computes pairwise similarity scores across all titles
5. Recommendation	Finds input title, ranks similar items, returns top-N results

---

📂 Project Structure

netflix-ml-system/
├── data/
│   └── cleaned_data.csv         # Preprocessed Netflix dataset
├── src/
│   ├── data_loader.py           # Loads and validates the dataset
│   ├── feature_builder.py       # Combines and weights text features
│   ├── vectorizer.py            # TF-IDF vectorization logic
│   ├── similarity.py            # Cosine similarity computation
│   └── recommender.py           # Recommendation interface
├── main.py                      # Entry point
├── requirements.txt
└── README.md

---

⚙️ Installation

Requirements: Python 3.8+

git clone <your-repo-url>
cd netflix-ml-system

python3 -m venv venv
source venv/bin/activate        # On Windows: venv\Scripts\activate

pip install -r requirements.txt

Dependencies:

pandas
numpy
scikit-learn

---

📦 Dataset

This project uses the Netflix Movies and TV Shows dataset, available on Kaggle.

Download netflix_titles.csv, run your cleaning script, and save the output as data/cleaned_data.csv.

The cleaned dataset should contain at minimum these columns:

Column	Description
title	Movie or show name
listed_in	Genre(s)
description	Short synopsis

---

▶️ Usage

python main.py

Example output:

Input: "Inception"

Top 5 Recommendations:
1. Interstellar        — Similarity: 0.87
2. The Dark Knight     — Similarity: 0.81
3. Tenet               — Similarity: 0.79
4. Shutter Island      — Similarity: 0.74
5. Memento             — Similarity: 0.71

---

🧪 Technologies Used

Library	Purpose
pandas	Data loading and manipulation
numpy	Numerical operations
scikit-learn	TF-IDF vectorization and cosine similarity

---

⚠️ Limitations

• Uses lexical (word-matching) similarity, not semantic understanding — "action" and "thriller" are treated as unrelated terms
• Feature weighting is manually tuned (heuristic), not learned from data
• No evaluation metrics (e.g., precision@k, recall) are implemented
• Cold-start problem: cannot recommend for titles not in the dataset

---

🔧 Future Improvements

• Semantic embeddings — replace TF-IDF with BERT or Sentence Transformers for contextual understanding
• TF-IDF tuning — experiment with n-grams, stopword removal, and min/max document frequency
• Evaluation pipeline — add offline metrics to measure recommendation quality
• Collaborative filtering — combine with user interaction data for hybrid recommendations
• API or web interface — expose recommendations via a REST API or simple UI

---

📌 Project Status

Milestone	Status
Data loading and cleaning	✅ Complete
Feature engineering	✅ Complete
TF-IDF vectorization	✅ Complete
Cosine similarity computation	✅ Complete
Recommendation pipeline	✅ Complete
Evaluation metrics	🔲 Planned
Web interface	🔲 Planned

---

🧠 Learning Outcomes

• Built an end-to-end ML pipeline from scratch with a modular architecture
• Applied TF-IDF vectorization and cosine similarity in a real-world context
• Practiced feature engineering with weighted text fields