Movie Recommender

A content-based movie recommender system that uses TF-IDF vectorization with custom weighting for numeric and text features (IMDB Rating, Meta Score, Genre, and Director). This project also includes a Streamlit-based frontend for a simple web interface.

Table of Contents

1. Project Overview
2. Features and Weights
3. Setup and Installation
4. Dataset
5. Running the Recommender (CLI)
6. Running the Streamlit App
7. Customization

---

Project Overview

This project demonstrates a content-based recommender system that reads a dataset of top 1000 IMDB movies, creates “weighted” textual features, and uses TF-IDF plus cosine similarity to find and recommend similar movies.

Core steps:

1. Data Loading: Reads the CSV containing movie metadata (titles, genres, directors, ratings, metascores, etc.).
2. Feature Weighting: Important numeric attributes (e.g., IMDB Rating, Meta Score) are converted to repeated tokens to give them heavier weight. Genre and Director are also repeated a certain number of times to reflect their relative importance.
3. TF-IDF Vectorization: Transforms the weighted text into a numerical vector.
4. Similarity: A cosine similarity matrix is computed for all movies.
5. Recommendation: Given a movie title, the system retrieves the top 10 similar movies.

---

Features and Weights

1. IMDB Rating (0–10) → repeated token counts equal to the rating (e.g., rating=8 → “rating” repeated 8 times).
2. Meta Score (0–100) → repeated token counts equal to (meta_score ÷ 10).
3. Genre → repeated 3× to give moderate-high weight.
4. Director → repeated 1× to give minimal weight.

These multipliers can be freely adjusted in code to emphasize or de-emphasize features.

---

Setup and Installation

1. Clone or Download: Copy this repo to your local machine.
2. Install Dependencies (ideally in a virtual environment):

   pip install pandas scikit-learn streamlit kagglehub