Using Metadata with Word2Vec to get recommendations in MovieLens

The idea of this post is to introduce an idea to use Word2Vec in a well-known dataset such as MovieLens. The approach is to define documents guided by what each user saw. After train we can relate films and its features. Also users can be related to them.

First we’ll get the data and sort by time. The reason of sorting by time is due to the fact that we need the movies that each user saw in chronological order.

And filter with ranking above of 3 points

Getting item features

One film can be in several genres at once. We can see this adding a column ‘Total’ that is the sum of all the genres.

In the following cell, we can see that the majority of the films have 2 features. In the next steps we will see why this is important.

Now we will create a dataset with all the users and all their watched films sorted by timestamp. Also we’ll add the first genre that is related to each movie. I know that a movie has several genres but this is to keep the example simple. Undoubtedly this could be coded better.

Therefore in the new dataset we can see the trace of watched films and its genre for each user. It’s time to use word2vec.

After train we have to see the results.

It seems that it works! The query of the word ‘Action’ is near to action movies such as ‘GoldenEye (1995)’. What about ‘Horror’?

All of the movies listed are about Horror. Thumbs up! :)

Let’s see movies related with Die Hard 2…

Let’s define a user as a vector of the average of movies that he/she saw.

Analysing the results we can see that almost the movies are comedies or family films. For instance ‘First Kid (1996)’ and ‘Father of the Bride Part II (1995)’ are definetly comedie films. Also the last one it’s romantic too. So, we can describe an user using their movies preferencies.

Thanks for reading! :)

Note: this notebook was originally posted in Kaggle Sept. 19 2018

Related posts:

Internet-driven Social media marketing Networking Websites are unique in their own personal utility and operations but of these namely Facebook, YouTube, etc. Instagram is special for the…

There is a parable at the beginning of the book, "And the Mountains Echoed" written by one of my favorite authors Khaled Hosseini, that led me to reflect on how people rationalize both their past…

Yes! on green rolling hills.. “Thespian Love.” is published by Allen Liao 廖晉德.