Movie-Recommender-System

Project completed on December 9, 2024.

Project Overview

This project is a Movie Recommender System that leverages collaborative filtering techniques to generate personalized movie recommendations. It is built using the MovieLens dataset, which includes approximately 1 million ratings from 6,040 users for 3,706 movies, providing a rich foundation for analyzing user preferences and movie similarities.

The recommender system is implemented from scratch, focusing on two key recommendation strategies:

1. System 1: Popularity-Based Recommendation – Recommends movies based on overall popularity.
2. System 2: Item-Based Collaborative Filtering (IBCF) – Uses similarity between movies to generate personalized recommendations for users.

Additionally, a Movie Recommender Web App was built to provide an interactive interface for users to explore and receive movie recommendations.

See Movie Recommender Web App demo below:

MovieRecommeder_demo.mov

Repository Contents

• Movie-Recommender-App/ - Folder containing the Dash web application that allows users to explore movie recommendations interactively.
  • app.py - The main Dash application that serves the web interface for recommendations.
  • myfuns.py - Contains helper functions for generating recommendations based on precomputed similarity scores.
  • popular_movies.csv - A dataset containing the most popular movies based on rating frequency and scores.
  • S_top_30.csv - A dataset containing similarity scores for the top 30 most similar movies for each title.
  • requirements.txt - A list of required Python libraries to run the Dash app.
• MovieImages/ - Folder containing movie posters used in the recommendations web app.
• ml-1m/ - Folder containing the MovieLens 1M dataset, which includes user ratings, movie metadata, and user demographic information.
  • movies.dat - Includes movie title, release year, and genres.
  • ratings.dat - 1,000,209 anonymous ratings from 6,040 users on 3,706 movies.
  • users.dat - Gender, age, occupation, and zip code.
• Instructions.pdf - Project details and problem statement.
• Movie-Recommender-System.ipynb - Jupyter Notebook containing the implementation of the recommender system, including data preprocessing, model development, and evaluation.
• Rmat.csv - User-movie rating matrix generated from the MovieLens dataset used as the input for IBCF.
• top_10_movies.csv - A list of the top 10 most recommended movies for each user based on Popularity-Based Recommendation.
• similarity_matrix.csv - Precomputed similarity scores between movies, used for IBCF. Download here

Methods and Techniques Used

1. Data Preprocessing

• Constructed a user-movie rating matrix from raw MovieLens ratings.
• Normalized ratings by centering each row to adjust for user rating biases.
• Removed movies with very few ratings to improve model efficiency.

2. Popularity-Based Recommendation

• Ranked movies based on the number of user ratings, considering only those with an average rating above 4.3.
• Recommended the top 10 most popular movies to all users.

3. Item-Based Collaborative Filtering (IBCF)

• Computed a movie similarity matrix using cosine similarity.
• Only retained the top 30 most similar movies for each title to enhance efficiency.
• Used similarity scores to generate personalized recommendations for users based on their watched movies.

4. Web Application

• Developed an interactive Dash web app where users can rate movies and receive recommendations.
• The app leverages precomputed similarity scores for fast, real-time recommendations.

How to Run the Web App

1. Install Dependencies:

pip install -r requirements.txt

2. Run the Dash App:

python app.py