Movie Cocktail App with Widgets Deployed with Binder
Movie Cocktail suggests new movies based on the selected combination of movies.
Not sure which movie to watch? Want to watch something similar to movies you've watched before?
Search and select movies you've already watch, mix them as you wish and get the closest movies to the cocktail.
Here is the link of the app:
How it Works?
Let's say you want to watch something 30% like 'the Matrix (1999)' and 70% like 'Titanic (1997)' then you can search for the movies the Matrix (1999) and Titanic (1997),combine them as you wish and find the top ten closest movies to this combination.
Note that, the program will work for a single movie as well, so you don't always have to mix the movies.
Distances?
These values are basically L2 distances between vectors. There are three different distance measures specified.
First distance on the top represents the average distance of the cocktail to any movie.
The second one represents the distance between the cocktail to the specified movie.
Extra distance info button shows the average distance of the specified movie to any other movie.
Under the Hood
Movie Cocktail uses movie embeddings obtained by a collaborative filtering model trained on 25m movie ratings.
By using the embedding vectors of selected movies, a new combined embedding vector is generated. After that, it is all about finding the closest L2 distances to the generated new embedding vector.
Tips
You don't have to mix the movies, you can just select a single movie, and find the closest movies to the specified movie. In this case slider value won't matter as long as it is greater than zero.
Search bar only allows searches of three or more characters. So if you are looking for a movie '21' you should search as '21 (' or if you're looking for the movie 'V' then you should search as 'V ('
This project is done for educational purposes only, there might be some bugs on the GUI.
