Movie Recommendation System

Project Overview

This project aims to build a robust recommendation system that offers various suggestions methods, based on the movies themselves as well as the audience's preferences, using the datasets from MovieLens and The Movies Dataset from Kaggle.

Main methodologies

• Text Processing with TF-IDF
• Machine Learning models: KNeighborsRegressor
• Statistical Methods: Jaccard similarity, Cosine similarity

Project Motivations:

Learning users' behaviors through their interactions with the products can be applied to promote relevant items strategically, thus leading to improvement in user satisfaction. Some of the state-of-art recommending systems have been widely utilized by big companies like Netflix, Amazon, etc.

This project is a collaborative effort of Winnie Nguyen and Fabio Turazzi, aiming to build a robust movie suggestion engine that takes advantage of a variety of available features. We use methods based on movie popularity, content filtering, collaborative filtering, and hybrid alternatives to maximize the relevancy to the recommendation output.

Recommendation Algorithms:

• Simple Recommendation System: generally analyzing most popular and highest rated movies across genres.

• Content-Based Recommendataion: using contents as the basis to suggest similar movies.

  • Genre similarity: recommendations based on the similarities between movie genres.
  • Description similarity: based on text similarities in movie descriptions.

• Collaborative Filtering: using ratings as the basis to suggest similar movies.

  • Item-based filtering: provides recommendations given a movie title by analyzing other items similarly rated by users.
  • User-based filtering: focuses on a target user to provide suggestions, by analyzing favorite movies by other similar users. Includes a model for rating prediction for a given user and movie.

Datasets

We use 4 datasets obtained from the 2 sources mentioned above.

• links.csv: containing movieId, imdbId, tmdbId (from MovieLens)
• movies.csv: containing movieId, title, genres (from MovieLens)
• ratings.csv: containing userId, movieId, rating, timestamp (from MovieLens)
• movies_metadata.csv: containing imdb_id, overview (from Kaggle with selected columns only)

links.csv

Variable name	Description	Data Type
movieId	unique Id given to a movie on MovieLens	int
imdbId	ununique Id given to a movie on IMDB.com	int
tmdbId	ununique Id given to a movie on themoviedb.org	float

movies.csv

Variable name	Description	Data Type
movieId	unique Id given to a movie	int64
title	movie title	object
genres	movie's genres	object

ratings.csv

Variable name	Description	Data Type
userId	unique ID for each user	int
movieId	unique ID for each movie on MovieLens	int
rating	rating given to the movie by user	float
timestamp	timestamp of given rating	int

movies_metadata.csv

Variable name	Description	Data Type
imdbId	ununique Id given to a movie on IMDB.com	object
overview	movie's overview	object

Project Directory

| -  recommendation_system        
|   -- data                                         Includes all datasets used for the project
|     --- links.csv                                 
|     --- movies.csv
|     --- ratings.csv                               Not uploaded due to file size constraints
|     --- movies_metadata.csv                       Not uploaded due to file size constraints
|     --- additional_datasets.md                    Describes the path for two datasets that couldn't be uploaded to GitHub 
|   -- LICENSE                                      MIT License
|   -- Movie_recommendation_sys.ipynb               Google Colab Notebook with source code
|   -- README.md                                    Project Overview

Potential Considerations

In this project, we were able to provide different perspectives to movie recommendation that can be used in a professional setting to generate a robust engine. Even though we were successfull in identifying item similarity, more methods can be explored to improve recommendations even further.