recomMe

A web application that uses Machine learning to fetch recommendations based on user's rating history.

Video at: https://www.youtube.com/watch?v=HfnsXtljXu0
Hosted at: https://recomme-engine.vercel.app/

• Note:
  
  • The backend is deployed on Heroku on a free account, so the application sleeps if not pinged every 30 minutes. Please wait after pressing register until you see a "Success" message/alert and only then proceed to login.
    
  • I would recommend cloning the repository and running it locally to get the full experience.

Datasets:

• Used 100k MovieLens Dataset, filtered it to use about 2300 movies and 100,000 user ratings.

Tech Stack and APIs:

Features:

1. Star rating: The user can rate movies on a scale of 0 to 5, which is saved in the database to recommend movies.
   
2. Cold-start problem: (when a new user has no history of ratings so recommendations can't be fetched)
   • Prevented by asking users to rate a few movies before fetching recommendations.
     
3. Content Based Recommendations: (refer to "content_based.py" and "get_content_based.py")
   • Uses cast, genres, director, and keywords as features/parameters.
   • Movies are found by applying count vectorization on the aggregated features and sorting movies based on cosine similarity.
     
4. Collaborative Filtering: (refer to "createSimilarityMatrix.py" and "collaborative_filtering.py" )
   
   • Capable of capturing the tastes of users and providing recommendations across genres.
   • Using the ratings of every movie, the Pearson correlation is calculated and stored to form an item-item similarity matrix.
   • The movies' ratings given by a new user are used to find similar movies by fetching the rows of the movie rated, from the similarity matrix and sorting it.
5. Search Engine: (refer to "createTableForSearch.py" and "getSearchResults.py")
   • The features/tags of the movies are aggregated and used to find the best match using a regular expression created from the search query.
   • This fetches and shows the movies with the highest "points" first.
   • "points" factor is calculated using movie popularity, vote_average, vote_count
   • Query is not limited to movie titles. Users can also search by genres/directors/cast/year of release or related keywords. ex: "Comedy movies", "Christopher Nolan movies", "Avatar"
6. Movie Info: Clicking on the movie card gives the user a brief about the movie.
   • Also displays movies similar to the particular movie clicked.
7. User Fact: Finds a keyword common in user's top-rated movies and displays it as a fact.
8. Wishlist: Users can save movies to their wishlist for later.
9. Refresh Recommendation Button: To get instant recommendations based on the movies rated.
10. Login System: Provided a fully responsive register/login system.
11. User Interface:
    • Provided an interactive user-friendly interface.
    • Fully optimized for smaller screens as well.

Installation Guide

• Make sure you have node and python installed.

1. Backend:

cd backend

install dependencies:

pip install -r requirements.txt

run server (running at port 5000):

python app.py

2. Frontend:

cd frontend

install dependencies:

npm i

run server (running at port 3000):

npm start