Movie Recommender System

A content-based movie recommendation system that suggests movies similar to a user's selection. The engine uses Natural Language Processing (NLP) to analyze movie metadata (genres, cast, crew, and keywords) and serves recommendations via a Streamlit web application.

🚀 Overview

This project processes the TMDB 5000 Movie Dataset to create a recommendation algorithm. Instead of using user ratings, it focuses on the content of the movies themselves.

• Data Processing: Cleans and merges datasets, extracts key features, and creates a unified "tag" system for every movie.
• Machine Learning: Uses CountVectorizer to convert text tags into vectors and calculates Cosine Similarity to find the closest matches in a 5000-dimensional space.
• Web App: A user-friendly interface built with Streamlit that displays movie recommendations and fetches real-time posters from the TMDB API.

🛠️ Technologies Used

• Python 3.x
• Pandas & NumPy: Data manipulation and analysis.
• Scikit-learn: Used for CountVectorizer and cosine_similarity.
• NLTK: Used PorterStemmer to reduce words to their root form (e.g., "dancing" → "danc").
• Streamlit: Frontend framework for the web application.
• TMDB API: Used to fetch movie posters dynamically.

📂 Project Structure

• movie-recommender-system.ipynb: Jupyter Notebook containing the data preprocessing pipeline, vectorization, and model generation.
• app.py: The main Streamlit application script.
• tmdb_5000_movies.csv: Metadata dataset (budget, overview, popularity, etc.).
• tmdb_5000_credits.csv: Credits dataset (cast, crew).
• movie.pkl: (Generated) Pickled dataframe containing movie titles and tags.
• similarity.pkl: (Generated) Pickled cosine similarity matrix.

⚙️ How It Works

1. Data Pipeline (movie-recommender-system.ipynb)

1. Merging: The movies and credits datasets are merged on the movie title.
2. Feature Extraction:
   • Genres & Keywords: Extracted from JSON format.
   • Cast: Top 3 actors are extracted.
   • Crew: The Director is isolated.
3. Text Cleaning: Spaces are removed from names (e.g., "Sam Worthington" becomes "SamWorthington") to create unique vector tokens.
4. Vectorization: A tags column is created by combining the overview, genres, keywords, cast, and crew. This text is stemmed and vectorized using a Bag-of-Words approach (5000 most frequent words).
5. Model Export: The resulting dataframe and similarity matrix are exported as .pkl files for the app to use.

2. The Application (app.py)

The app loads the pre-trained models and provides a dropdown menu for movie selection. When the "Recommend" button is clicked, the system:

1. Finds the index of the selected movie.
2. Retrieves the 5 most similar movies based on the cosine similarity matrix.
3. Fetches poster URLs using the TMDB API.
4. Displays the titles and posters in a 5-column grid.

🔧 Setup & Installation

1. Clone the repository:

   git clone [https://github.com/yourusername/movie-recommender-system.git](https://github.com/yourusername/movie-recommender-system.git)

2. Install dependencies:

   pip install streamlit pandas numpy scikit-learn nltk requests

3. Generate Models: Run the Jupyter Notebook to generate the necessary pickle files. Open movie-recommender-system.ipynb in Jupyter and run all cells. This will create movie.pkl and similarity.pkl.

4. Run the App:

   streamlit run app.py

📝 API Configuration

The app.py file contains an authorization bearer token for the TMDB API to fetch posters.