Skip to content

Latest commit

 

History

History
129 lines (76 loc) · 4.41 KB

README.md

File metadata and controls

129 lines (76 loc) · 4.41 KB

A one-stop shop for movie recommendation, integrating multiple recommendation algorithms tested and verified on real users for optimised results.

Development Timeline

  • Week 1 - Ideation and Initial Documentation
  • Week 2 - Backend Development
  • Week 3 - Frontend Development
  • Week 4 - Final Touches, Integration and Submission

Tech Stack

Client: Streamlit

Server: Numpy, Pandas, Scikit-Learn

Run Locally

Clone the project

  git clone git@github.com:abhishekg495/RecommenderSystem.git

Go to the project directory

  cd RecommenderSystem

Install dependencies

  pip install -r requirements.txt

Start the server

  streamlit run main.py

Folder Structure

  • Datasets - contains all the required data in csv format.

    Webscraping and data processing has been used to produce required information from exisiting datasets on Kaggle.

    • average_ratings.csv
    • links.csv
    • movies_features.csv
    • ratings_sorted_movies.csv
    • ratings.csv

  • JupyterNotebooks - contains Python code in Jupyter notebooks format.

    These notebooks are not required for the app to run, but are made available for reference as all the data processing and webscraping was done using Jupyter notebook environment.

    • BasicRecommender
    • CollaborativeFiltering
    • ContentBasedFiltering

  • Posters - contains posters for all movies, along with website favicon and other required images.

    Movies' posters have been web scraped from IMDB using Python.

  • Screenshots - contains screenshots and demo of the deployed project webpage.

  • basic_recommender.py collaborative_filtering.py content_based_filtering.py - contain Python classes for different recommendation algorithms.

  • basic_ui.py collaborative_ui.py content_based_ui.py - contain Python classes for frontend elements for different recommendation algorithms.

  • main.py - driver script to run the application.

  • posters_printer.py - contains Python class to render posters for a given list of movies, along with their IMDB lnks.

  • watchlist.py - contains Python class to mantain a watchlist of the user's chosen movies.

Features

  • Sort movies based on ratings from thousands of users
  • Select your favorite movies to find similar titles
  • Search for movies using keywords (actors, tags, directors etc.)
  • Get personalised recommendations based on your own ratings
  • Automatic light/dark mode toggle based on device theme
  • Easy-to-use, smooth and responsive UI
  • Cross platform
  • Add movies to your watchlist
  • Accurate recommendations

Demo

Click here for the video demo

Screenshots

Sort by Ratings/Popularity (and choose genres)

App Screenshot 1

Search for a movie based on its content

App Screenshot 2

App Screenshot 3

Get personalised recommendations based on your own ratings

App Screenshot 4

Lessons Learned

The project started out with a simple inefficient implementation of the ratings/popularity based sorting feature. This taught me how to sort information stored in a DataFrame according to different conditions.

The next part was the content-based searching feature. I had to go through a whole bunch of different metrics and types of NLP models to get the most accurate results that I could.

The third feature was implemented in the form of personalised recommendations based on a user's ratings for different movies. Several similarity metrics had to be tried and tested to implement an approach similar to KNN but in a much more elegant and simplified manner.

Last but not the least, a major part of the development cycle was spent in making the app more user friendly in general and in using several optimisation techniques like caching to make the experience smoother for the user. Web scraping also played a major role as gathering accurate data was a much more crucial and tedious task as compared to implementing ML models based on that data.