Movie-Recommender

Live: https://what-should-i-watch-tonight.herokuapp.com/

• Need someone to just recommend one movie to you without having the hassle to go through ten different streaming apps?
• This recommendation system is designed to minimize the users overthinking and contemplation.
• We will give you one perfect choice so that you dont have to spend hours searching for what you like.
• What if you dont like the movie we recommended? No worries, just press pass and we will provide you with another movie. It is that simple.

Run app.py to access the website locally

How it works

• Choose the genre you are in the mood for.
• You can also specify the language and the preferable release date of the movie you wish to watch.
• Then, select the movies you have watched and liked from a list of 10 movies picked from the top 250 following the set conditions.
• If you do not like any of the 10 options provided you can always press refresh and ask for more.
• Depending on the choices you make we provide you with one movie perfectly suited for you.
• If you are not satisfied with the recommendation, no worries, press pass and we will recommend another movie for you.

Recommendation System

• We use the Content-Based Recommendation system where we take your previous likings in order to recommend a movie to you.
• Similar movies are found by transforming parameters like plot, keywords, director, writer, actor, and genre to vectors and then using cosine similarity on them.
• We use sklearn CountVectorizer to convert the parameters from text to vector/token counts.
• Cosine Similarity is used to plot the similarities. The output varies from 0-1 where where 1 denotes 100% similarity.
• Using this we can find the cosΘ between the two vectors.
• We then sort the possible recommendations for each movie based on the number of times it was suggested primarily and the movie’s rating to find the perfect match.

Dataset

• We got ratings, crew information, genre for every movie on iMDb from https://www.imdb.com/interfaces/.
• We then got the poster, plot, original language and keywords for every movie using the unique iMDb id with the help of TMDB API
• We removed all movies pre 1990 for relevance.
• We then grouped all the movies by genre and rated them using the True Bayesian estimation.
• Weighted Rating(WR) = (v/v+m).R + (m/m+v).C
  • Where v is the number of votes, m is the minimum number of votes the movie should get to be considered, C is mean vote and R is the average rating of the movie.
  • We chose m to be 80th percentile of the number of votes in the database which turns out to be 4229 votes.
  • C for our dataset was around 6.