WatchNext- A recommendation website with a twist
Do you love to watch movies? Ever felt that your recommendations on Netflix or similar streaming services have become boring? I present to you “WatchNext”. It’s a movie recommendation website, but with a lot of differences from the existing ones. How? Well that’s what the blog is about.
First let’s talk about the kind of problems that you face with regular recommendation engines which you’ll find online and the ones on streaming sites.
Most movies are not available on all streaming platforms. Unless you’re super rich or have something like the Amazon Fire TV Stick, I’m sure you don’t have an account in all of them. Most of us don’t even have a single account.
The other problem is that most of the popular, most liked and collaborative recommendations are based on a simple ‘thumbs up’ / ’thumbs down’ or on a numerical rating. We really can’t determine which movie is better based on the number of thumbs up it got.
A Reddit user accurately said, “The problem with the thumbs up / thumbs down is that there is now absolutely no difference between I guess I didn’t hate this movie and This is the greatest movie I’ve ever seen in my life, and I want to see more like it “.
It is also highly probable that the actor or director had a huge fan base to watch that movie and its pretty obvious that they’ll give it a thumbs up no matter how bad the story was. The problem with numeric rating is that its relative. What might be a 7 rating to one user might be a 5 to me. It depends on the expectations the person has. It’s also really hard and confusing to rate a movie with numbers.
The other problem with common recommendation systems is that they do not usually recommend the less popular or mainstream movies.
“Documentaries typically struggle to gain as much attention as action and fantasy shows such as Game of Thrones, and the same is true of historical dramas,” The Economist wrote.
The same goes for movies too. Almost all of us don’t know about the amazing movies with the best plots out there, simply because it wasn’t mainstream, didn’t involve a famous actor/director and most importantly because Netflix didn’t recommend it.
Now that I’ve stated out the problems, this is how my team and I plan to solve them.
The website will ask users what all streaming services they have access to, and based on the choices, the movies will be recommended accordingly. Don’t fret if you don’t have any account, there’s always Torrent (the ultimate database for all movies *wink*) which we’ve included in the options. So now you will only be recommended movies that you can access.
To deal with the rating problem, we apply sentiment analysis to IMDb movie reviews and reviews from various other websites for each movie. The scores of average polarity of all reviews (how bad or good a review is) for each movie is added to the movie rating for a normalized score. This movie score will be used to rank them.
To include the less known movies, there will a be list where movies that has less viewers and a higher movie score will be recommended to the user. This will be based on the genre and the type of movies the user watches, or can be completely random, just as the user wishes.
The remaining part of the recommendation website will work like the usual ones. Movies will be recommended based on popularity (based on most number of watches and movie score), collaborative and content based filtering. An option to include language and age preferences will also be present. Content based filtering will be based on the plot, director and the main actor and will be done using cosine similarity. Collaborative filtering will be based on the likes of similar users who will be determined using similar watch patterns, age, region etc.
With that, I am concluding this blog. The website is still under progress. Once that is done, I will include the technical details about the models, datasets and code in my next blog. Till then farewell!
